US20200133308A1

Movatterモバイル変換

Info

Publication number: US20200133308A1
Application number: US16/729,320
Authority: US
Inventors: Igal RAICHELGAUZ; Karina ODINAEV
Original assignee: Cartica AI Ltd
Current assignee: Autobrains Technologies Ltd
Priority date: 2018-10-18
Filing date: 2019-12-28
Publication date: 2020-04-30
Also published as: US20200258390A1; US20200133266A1; US20200255004A1; US11029685B2; US20200135029A1; US11718322B2; US20200258391A1; US12415547B2; US11673583B2; US12128927B2; US20200130684A1; US11685400B2; US20200130708A1; US11087628B2; US20200218257A1

Abstract

V2V communication-less truck platooning.

Description

CROSS REFERENCE

This application is a continuation of PCT/IB2019/058207 filing date Sep. 27, 2019 which in turn claims priority from U.S. provisional Ser. No. 62/747,147 filing date Oct. 18 2018.

This application claims priority from U.S. provisional patent Ser. No. 62/827,112 filing date Mar. 31, 2019, both patent applications are incorporated herein by reference.

TECHNICAL FIELD

The present disclosure generally relates to detecting and avoiding obstacles in an autonomous driving environment.

BACKGROUND

Truck platooning is the linking of two or more trucks in convoy, heavily based on connectivity technology and automated driving support systems. These vehicles automatically maintain a set, close distance between each other when they are connected for certain parts of a journey, for instance on motorways.

Truck platooning holds great potential to make road transport safer, cleaner and more efficient in the future.

Platooning results in a lower fuel consumption, as the trucks drive closer together at a constant speed, with less braking and accelerating.

Truck platooning has the potential to reduce CO2 emissions by up to 10%. —With conventional trucks, critical risk factors are driver reaction time and concentration. Indeed, some 90% of all traffic accidents are due to human error.

Allowing for more predictive driving of trucks on the road, platooning also improves safety for other road users.

Platooning is a cost-saver, as lower fuel consumption means lower fuel costs, which currently make up 30% of total operating costs of a truck.

Truck platooning is based on communicating between trucks using propriety vehicle to vehicle (V2V) communication protocols and thus is limited to performing truck platooning of trucks that are operated by the same company that leave the facility together and travel the same route till the end of the journey.

The V2V communication is used to convey commands from the front truck to the autonomous driving models that autonomously drive the trucks that blindly follow the front truck.

The reliance on the V2V communication may endanger the trucks of the platoon and their environment in cases where the V2V communication is interrupted or jammed and in cases when an error of the front truck causes other trucks to follow it and also suffer from the error.

There is a growing need to provide a safer and more flexible method of truck platooning.

SUMMARY

There may be provided a method for vehicle to vehicle communication less truck platooning, the method may include monitoring, by a truck monitor, an environment of the truck to provide monitoring results; determining, based on the sensed information, whether to engage in truck platooning with a set of at least one other truck; engaging the truck platooning following determination to engage in the truck platooning; wherein the engaging comprises at least partially autonomously driving the truck based on the monitoring results.

The method may include determining, based at least on the monitoring results, whether disengage the truck platooning.

At least one of the determining of whether to engage and whether to disengage may be based on a risk level associated with the truck platooning.

At least one of the determining of whether to engage and whether to disengage may be based on a comparison between the path of progress of the set and a desired path of progress of the truck.

At least one of the determining of whether to engage and whether to disengage may be based on an assessment of a risk level associated with the progress of the set.

At least one of the determining of whether to engage and whether to disengage may be based on an assessment of a risk level associated with road conditions.

At least one of the determining of whether to engage and whether to disengage may be based on an assessment of a risk level imposed by an entrance of a vehicle between members of a convoy.

At least one of the determining of whether to engage and whether to disengage may be based on a physiological state of a driver of the truck.

At least one of the determining of whether to engage and whether to disengage may be based on at least one out of (a) a relationship between a type of the vehicle and a type of the certain truck, (b) a relationship between a size of the vehicle and a size of the certain truck, and (c) a relationship between a weight of the vehicle and a weight of the certain truck.

The engaging the truck platooning comprises engaging as a leading truck.

The engaging the truck platooning comprises engaging as a following truck.

The engaging comprises engaging the platooning at certain location, and then replacing locations with at least one other truck.

The method may include receiving commands conveyed over vehicle to vehicle communication, wherein the commands may be for determining driving of trucks while engaged in the truck platooning.

The at least partially autonomously driving the truck may be executed regardless of the commands.

The method may include analyzing an accuracy of the commands and generating an alerts when the commands may be inaccurate.

The method may include analyzing an accuracy of the commands and disengaging when the commands may be inaccurate.

The method may include dynamically determining am amount of control of the vehicle during the at least partially autonomously driving of the truck.

There may be provided a non-transitory computer readable medium for vehicle to vehicle communication less truck platooning, the non-transitory computer readable medium stores instructions for: monitoring by a truck monitor, an environment of the truck to provide monitoring results; determining, based on the sensed information, whether to engage in truck platooning with a set of at least one other truck; engaging the truck platooning following determination to engage in the truck platooning; wherein the engaging comprises at least partially autonomously driving the truck based on the monitoring results.

The engaging the truck platooning comprises engaging as a leading truck.

The engaging the truck platooning comprises engaging as a following truck.

There may be provided a system for vehicle to vehicle communication less truck platooning, the system may include a truck monitor that may be configured to: (i) monitor an environment of the truck to provide monitoring results; (ii) determine, based on the sensed information, whether to engage in truck platooning with a set of at least one other truck; (iii) and participate in engaging the truck platooning following determination to engage in the truck platooning; wherein the engaging comprises at least partially autonomously driving the truck based on the monitoring results.

The truck monitor may include a processor and a memory unit. The truck monitor may include one or more sensors or may receive sensed information from one or more sensors. The truck monitor may include a module for controlling the at least partially autonomously driving of the truck and/or may communicate and/or send requests and/or commands to such module. The participation may include sensing a request, triggering, sending a command, providing information, partially or fully executing the at least partially autonomously driving the truck.

BRIEF DESCRIPTION OF THE DRAWINGS

The embodiments of the disclosure will be understood and appreciated more fully from the following detailed description, taken in conjunction with the drawings in which:

FIG. 1A illustrates an example of a method;

FIG. 1B illustrates an example of a signature;

FIG. 1C illustrates an example of a dimension expansion process;

FIG. 1D illustrates an example of a merge operation;

FIG. 1E illustrates an example of hybrid process;

FIG. 1F illustrates an example of a first iteration of the dimension expansion process;

FIG. 1G illustrates an example of a method;

FIG. 1H illustrates an example of a method;

FIG. 1I illustrates an example of a method;

FIG. 1J illustrates an example of a method;

FIG. 1K illustrates an example of a method;

FIG. 1L illustrates an example of a method;

FIG. 1M illustrates an example of a method;

FIG. 1N illustrates an example of a matching process and a generation of a higher accuracy shape information;

FIG. 1O illustrates an example of an image and image identifiers;

FIG. 1P illustrates an example of an image, approximated regions of interest, compressed shape information and image identifiers;

FIG. 1Q illustrates an example of an image, approximated regions of interest, compressed shape information and image identifiers;

FIG. 1R illustrates an example of an image;

FIG. 1S illustrates an example of a method;

FIG. 2A illustrates an example of images of different scales;

FIG. 2B illustrates an example of images of different scales;

FIG. 2C illustrates an example of a method;

FIG. 2D illustrates an example of a method;

FIG. 2E illustrates an example of a method;

FIG. 2F illustrates an example of a method;

FIG. 2G illustrates an example of different images;

FIG. 2H illustrates an example of a method;

FIG. 2I illustrates an example of a method;

FIG. 2J illustrates an example of a method;

FIG. 2K illustrates an example of different images acquisition angles;

FIG. 2L illustrates an example of a method;

FIG. 2M illustrates an example of a method;

FIG. 2N illustrates an example of a system;

FIG. 3A is a partly-pictorial, partly-block diagram illustration of an exemplary obstacle detection and mapping system, constructed and operative in accordance with embodiments described herein;

FIG. 3B is a block diagram of an exemplary autonomous driving system to be integrated in the vehicle ofFIG. 3A;

FIG. 3C is a flowchart of an exemplary process to be performed by the autonomous driving system ofFIG. 3B;

FIG. 4 is a block diagram of an exemplary obstacle avoidance server ofFIG. 3A;

FIG. 5 is a flowchart of an exemplary process to be performed by the obstacle avoidance server ofFIG. 4;

FIG. 6 is an example of a method;

FIG. 7 is an example of a method;

FIG. 8 is an example of a driving scenario;

FIG. 9 is an example of a driving scenario;

FIG. 10 is an example of a driving scenario;

FIG. 11 is an example of a method;

FIG. 12 is an example of a method;

FIG. 13 is an example of a method;

FIG. 14 is an example of a driving scenario;

FIG. 15 is an example of a driving scenario;

FIG. 16 is an example of a scene;

FIG. 17 is an example of a scene;

FIG. 18 is an example of a driving scenario;

FIG. 19 is an example of a method;

FIG. 20 is an example of a method;

FIG. 21 is an example of a method;

FIG. 22 is an example of a driving scenario;

FIG. 23 is an example of a driving scenario;

FIG. 24 is an example of a driving scenario;

FIG. 25 is an example of a driving scenario;

FIG. 26 is an example of a driving scenario;

FIG. 27 is an example of a driving scenario;

FIG. 28 is an example of a method;

FIG. 29 is an example of entity movement functions;

FIG. 30 is an example of a method;

FIG. 31 is an example of a method;

FIG. 32 is an example of a method;

FIG. 33 is an example of a method;

FIG. 34 is an example of a method;

FIG. 35 is an example of a method;

FIG. 36 is an example of a driving scenario;

FIG. 37 is an example of a method;

FIG. 38 is an example of a method;

FIGS. 39-44 illustrate various data structures including a concept, test images and matching results, as well as various processes related to the data structures;

FIG. 45 is an example of a method;

FIG. 46 is an example of a method;

FIG. 47 is a timing diagram;

FIG. 48 is an example of a method;

FIG. 49 is an example of a method;

FIG. 50 is an example of a method;

FIG. 51 is an example of a method;

FIG. 52 is an example of a method;

FIG. 53 is an example of an image;

FIG. 54 is an example of an image;

FIG. 55 is an example of an image;

FIG. 56 illustrates examples of two signatures of different resolutions;

FIG. 57 illustrates an example of various data structures and processes;

FIG. 58 illustrates an example of a vehicle and a system;

FIG. 59 is an example of a method; and

FIG. 60 is an example of a method.

DESCRIPTION OF EXAMPLE EMBODIMENTS

The specification and/or drawings may refer to an image. An image is an example of a media unit. Any reference to an image may be applied mutatis mutandis to a media unit. A media unit may be an example of sensed information. Any reference to a media unit may be applied mutatis mutandis to ay type of natural signal such as but not limited to signal generated by nature, signal representing human behavior, signal representing operations related to the stock market, a medical signal, financial series, geodetic signals, geophysical, chemical, molecular, textual and numerical signals, time series, and the like. Any reference to a media unit may be applied mutatis mutandis to sensed information. The sensed information may be of any kind and may be sensed by any type of sensors—such as a visual light camera, an audio sensor, a sensor that may sense infrared, radar imagery, ultrasound, electro-optics, radiography, LIDAR (light detection and ranging), etc. The sensing may include generating samples (for example, pixel, audio signals) that represent the signal that was transmitted, or otherwise reach the sensor.

The specification and/or drawings may refer to a spanning element. A spanning element may be implemented in software or hardware. Different spanning element of a certain iteration are configured to apply different mathematical functions on the input they receive. Non-limiting examples of the mathematical functions include filtering, although other functions may be applied.

The specification and/or drawings may refer to a concept structure. A concept structure may include one or more clusters. Each cluster may include signatures and related metadata. Each reference to one or more clusters may be applicable to a reference to a concept structure.

The specification and/or drawings may refer to a processor. The processor may be a processing circuitry. The processing circuitry may be implemented as a central processing unit (CPU), and/or one or more other integrated circuits such as application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), full-custom integrated circuits, etc., or a combination of such integrated circuits.

Any combination of any steps of any method illustrated in the specification and/or drawings may be provided.

Any combination of any subject matter of any of claims may be provided.

Any combinations of systems, units, components, processors, sensors, illustrated in the specification and/or drawings may be provided.

Low Power Generation of Signatures

The analysis of content of a media unit may be executed by generating a signature of the media unit and by comparing the signature to reference signatures. The reference signatures may be arranged in one or more concept structures or may be arranged in any other manner. The signatures may be used for object detection or for any other use.

The signature may be generated by creating a multidimensional representation of the media unit. The multidimensional representation of the media unit may have a very large number of dimensions. The high number of dimensions may guarantee that the multidimensional representation of different media units that include different objects is sparse—and that object identifiers of different objects are distant from each other—thus improving the robustness of the signatures.

The generation of the signature is executed in an iterative manner that includes multiple iterations, each iteration may include an expansion operations that is followed by a merge operation. The expansion operation of an iteration is performed by spanning elements of that iteration. By determining, per iteration, which spanning elements (of that iteration) are relevant—and reducing the power consumption of irrelevant spanning elements—a significant amount of power may be saved.

In many cases, most of the spanning elements of an iteration are irrelevant—thus after determining (by the spanning elements) their relevancy—the spanning elements that are deemed to be irrelevant may be shut down a/or enter an idle mode.

FIG. 1A illustrates amethod5000 for generating a signature of a media unit.

Method

5000 may start bystep5010 of receiving or generating sensed information.

The sensed information may be a media unit of multiple objects.

Step

5010 may be followed by processing the media unit by performing multiple iterations, wherein at least some of the multiple iterations comprises applying, by spanning elements of the iteration, dimension expansion process that are followed by a merge operation.

The processing may include:

Step

5020 of performing a k′th iteration expansion process (k may be a variable that is used to track the number of iterations).

Step

5030 of performing a k′th iteration merge process.

Step

5040 of changing the value of k.

Step

5050 of checking if all required iterations were done—if so proceeding to step5060 of completing the generation of the signature. Else—jumping to step5020.

The output ofstep5020 is a k′th iteration expansion results5120.

The output ofstep5030 is a k′th iteration merge results5130.

For each iteration (except the first iteration)—the merge result of the previous iteration is an input to the current iteration expansion process.

At least some of the K iterations involve selectively reducing the power consumption of some spanning elements (during step5020) that are deemed to be irrelevant.

FIG. 1B is an example of animage signature6027 of a media unit that is an image6000 and of anoutcome6013 of the last (K′th) iteration.

Theimage6001 is virtually segments to segments6000(i,k). The segments may be of the same shape and size, but this is not necessarily so.

Outcome

6013 may be a tensor that includes a vector of values per each segment of the media unit. One or more objects may appear in a certain segment. For each object—an object identifier (of the signature) points to locations of significant values, within a certain vector associated with the certain segment.

For example—a top left segment (6001(1,1)) of the image may be represented in theoutcome6013 by a vector V(1,1)6017(1,1) that has multiple values. The number of values per vector may exceed 100, 200, 500, 1000, and the like.

The significant values (for example—more than 10, 20, 30, 40 values, and/or more than 0.1%, 0.2%. 0.5%, 1%, 5% of all values of the vector and the like) may be selected. The significant values may have the values—but may be selected in any other manner.

FIG. 1B illustrates a set of significant responses6015(1,1) of vector V(1,1)6017(1,1). The set includes five significant values (such as first significant value SV1(1,1)6013(1,1,1), second significant value SV2(1,1), third significant value SV3(1,1), fourth significant value SV4(1,1), and fifth significant value SV5(1,1)6013(1,1,5).

Theimage signature6027 includes five indexes for the retrieval of the five significant values—first till fifth identifiers ID1-ID5 are indexes for retrieving the first till fifth significant values.

FIG. 1C illustrates a k′th iteration expansion process.

The k′th iteration expansion process start by receiving themerge results5060′ of a previous iteration.

The merge results of a previous iteration may include values are indicative of previous expansion processes—for example—may include values that are indicative of relevant spanning elements from a previous expansion operation, values indicative of relevant regions of interest in a multidimensional representation of the merge results of a previous iteration.

The merge results (of the previous iteration) are fed to spanning elements such as spanning elements5061(1)-5061(J).

Each spanning element is associated with a unique set of values. The set may include one or more values. The spanning elements apply different functions that may be orthogonal to each other. Using non-orthogonal functions may increase the number of spanning elements—but this increment may be tolerable.

The spanning elements may apply functions that are decorrelated to each other—even if not orthogonal to each other.

The spanning elements may be associated with different combinations of object identifiers that may “cover” multiple possible media units. Candidates for combinations of object identifiers may be selected in various manners—for example based on their occurrence in various images (such as test images) randomly, pseudo randomly, according to some rules and the like. Out of these candidates the combinations may be selected to be decorrelated, to cover said multiple possible media units and/or in a manner that certain objects are mapped to the same spanning elements.

Each spanning element compares the values of the merge results to the unique set (associated with the spanning element) and if there is a match—then the spanning element is deemed to be relevant. If so—the spanning element completes the expansion operation.

If there is no match—the spanning element is deemed to be irrelevant and enters a low power mode. The low power mode may also be referred to as an idle mode, a standby mode, and the like. The low power mode is termed low power because the power consumption of an irrelevant spanning element is lower than the power consumption of a relevant spanning element.

InFIG. 1C various spanning elements are relevant (5061(1)-5061(3)) and one spanning element is irrelevant (5061(J)).

Each relevant spanning element may perform a spanning operation that includes assigning an output value that is indicative of an identity of the relevant spanning elements of the iteration. The output value may also be indicative of identities of previous relevant spanning elements (from previous iterations).

For example—assuming that spanning element number fifty is relevant and is associated with a unique set of values of eight and four—then the output value may reflect the numbers fifty, four and eight—for example one thousand multiplied by (fifty+forty) plus forty. Any other mapping function may be applied.

FIG. 1C also illustrates the steps executed by each spanning element:

Checking if the merge results are relevant to the spanning element (step5091).

If-so—completing the spanning operation (step5093).

If not—entering an idle state (step5092).

FIG. 1D is an example of various merge operations.

A merge operation may include finding regions of interest. The regions of interest are regions within a multidimensional representation of the sensed information. A region of interest may exhibit a more significant response (for example a stronger, higher intensity response).

The merge operation (executed during a k′th iteration merge operation) may include at least one of the following:

Step S031 of searching for overlaps between regions of interest (of the k′th iteration expansion operation results) and define regions of interest that are related to the overlaps.

Step

5032 of determining to drop one or more region of interest, and dropping according to the determination.

Step

5033 of searching for relationships between regions of interest (of the k′th iteration expansion operation results) and define regions of interest that are related to the relationship.

Step

5034 of searching for proximate regions of interest (of the k′th iteration expansion operation results) and define regions of interest that are related to the proximity. Proximate may be a distance that is a certain fraction (for example less than 1%) of the multi-dimensional space, may be a certain fraction of at least one of the regions of interest that are tested for proximity.

Step

5035 of searching for relationships between regions of interest (of the k′th iteration expansion operation results) and define regions of interest that are related to the relationship.

Step5036 of merging and/or dropping k′th iteration regions of interest based on shape information related to shape of the k′th iteration regions of interest.

The same merge operations may be applied in different iterations.

Alternatively, different merge operations may be executed during different iterations.

FIG. 1E illustrates an example of a hybrid process and aninput image6001.

The hybrid process is hybrid in the sense that some expansion and merge operations are executed by a convolutional neural network (CNN) and some expansion and merge operations (denoted additional iterations of expansion and merge) are not executed by the CNN—but rather by a process that may include determining a relevancy of spanning elements and entering irrelevant spanning elements to a low power mode.

InFIG. 1E one or more initial iterations are executed by first and second CNN layers6010(1) and6010(2) that apply first and second functions6015(1) and6015(2).

The output of these layers provided information about image properties. The image properties may not amount to object detection. Image properties may include location of edges, properties of curves, and the like.

The CNN may include additional layers (for example third till N'th layer6010(N)) that may provide aCNN output6018 that may include object detection information. It should be noted that the additional layers may not be included.

It should be noted that executing the entire signature generation process by a hardware CNN of fixed connectivity may have a higher power consumption—as the CNN will not be able to reduce the power consumption of irrelevant nodes.

FIG. 1F illustrates aninput image6001, and a single iteration of an expansion operation and a merge operation.

InFIG. 1F theinput image6001 undergoes two expansion operations.

The first expansion operation involves filtering the input image by afirst filtering operation6031 to provide first regions of interest (denoted1) in a firstfiltered image6031′.

The first expansion operation also involves filtering the input image by asecond filtering operation6032 to provide first regions of interest (denoted2) in a secondfiltered image6032′,

The merge operation includes merging the two images by overlaying the first filtered image on the second filtered image to provide regions of

interest

1,2,12 and21. Region ofinterest12 is an overlap area shared by a certain region ofinterest1 and a certain region ofinterest2. Region ofinterest21 is a union of another region ofinterest1 and another region ofinterest2.

FIG. 1G illustratesmethod5200 for generating a signature.

Method

5200 may include the following sequence of steps:

Step

5210 of receiving or generating an image.

Step

5220 of performing a first iteration expansion operation (which is an expansion operation that is executed during a first iteration)

Step

5230 of performing a first iteration merge operation.

Step

5240 of amending index k (k is an iteration counter). InFIG. 7 in incremented by one—this is only an example of how the number of iterations are tracked.

Step

5260 of performing a k′th iteration expansion operation on the (k-1)′th iteration merge results.

Step

5270 of performing a k′th iteration merge operation (on the k′th iteration expansion operation results.

Step

5280 of changing the value of index k.

Step

5290 of checking if all iteration ended (k reached its final value—for example K).

If no—there are still iterations to be executed—jumping fromstep5290 to step S260.

If yes—jumping to step5060 of completing the generation of the signature. This may include, for example, selecting significant attributes, determining retrieval information (for example indexes) that point to the selected significant attributes.

Step

5220 may include:

Step

5222 of generating multiple representations of the image within a multi-dimensional space of f(1) dimensions. The expansion operation ofstep5220 generates a first iteration multidimensional representation of the first image. The number of dimensions of this first iteration multidimensional representation is denoted f(1).

Step

5224 of assigning a unique index for each region of interest within the multiple representations. For example, referring toFIG. 6—indexes1 andindexes2 are assigned to regions of interests generated during the first

iteration expansion operations

6031 and6032.

Step

5230 may include:

Step

5232 of searching for relationships between regions of interest and define regions of interest that are related to the relationships. For example—union or intersection illustrate dinFIG. 6.

Step

5234 of assigning a unique index for each region of interest within the multiple representations. For example—referring toFIG. 6—

indexes

1,2,12 and21.

Step

5260 may include:

Step

5262 of generating multiple representations of the merge results of the (k-1)′th iteration within a multi-dimensional space of f(k) dimensions. The expansion operation ofstep5260 generates a k′th iteration multidimensional representation of the first image. The number of dimensions of this kth iteration multidimensional representation is denoted f(k).

Step

5264 of assigning a unique index for each region of interest within the multiple representations.

Step

5270 may include

Step

5272 of searching for relationships between regions of interest and define regions of interest that are related to the relationships.

Step

5274 of Assigning a unique index for each region of interest within the multiple representations.

FIG. 1H illustrates amethod5201. Inmethod5201 the relationships between the regions of interest are overlaps.

Thus—step5232 is replaced bystep5232′ of searching for overlaps between regions of interest and define regions of interest that are related to the overlaps.

Step

5272 is replaced bystep5272′ of searching for overlaps between regions of interest and define regions of interest that are related to the overlaps.

FIG. 11 illustrates amethod7000 for low-power calculation of a signature.

Method

7000 starts bystep7010 of receiving or generating a media unit of multiple objects.

Step

7010 may be followed bystep7012 of processing the media unit by performing multiple iterations, wherein at least some of the multiple iterations comprises applying, by spanning elements of the iteration, dimension expansion process that are followed by a merge operation.

The applying of the dimension expansion process of an iteration may include (a) determining a relevancy of the spanning elements of the iteration; and (b) completing the dimension expansion process by relevant spanning elements of the iteration and reducing a power consumption of irrelevant spanning elements until, at least, a completion of the applying of the dimension expansion process.

The identifiers may be retrieval information for retrieving the significant portions.

The at least some of the multiple iterations may be a majority of the multiple iterations.

The output of the multiple iteration may include multiple property attributes for each segment out of multiple segments of the media unit; and wherein the significant portions of an output of the multiple iterations may include more impactful property attributes.

The first iteration of the multiple iteration may include applying the dimension expansion process by applying different filters on the media unit.

The at least some of the multiple iteration exclude at least a first iteration of the multiple iterations. See, for example,FIG. 1E.

The determining the relevancy of the spanning elements of the iteration may be based on at least some identities of relevant spanning elements of at least one previous iteration.

The determining the relevancy of the spanning elements of the iteration may be based on at least some identities of relevant spanning elements of at least one previous iteration that preceded the iteration.

The determining the relevancy of the spanning elements of the iteration may be based on properties of the media unit.

The determining the relevancy of the spanning elements of the iteration may be performed by the spanning elements of the iteration.

Method

7000 may include a neural network processing operation that may be executed by one or more layers of a neural network and does not belong to the at least some of the multiple iterations. See, for example,FIG. 1E.

The at least one iteration may be executed without reducing power consumption of irrelevant neurons of the one or more layers.

The one or more layers may output information about properties of the media unit, wherein the information differs from a recognition of the multiple objects.

The applying, by spanning elements of an iteration that differs from the first iteration, the dimension expansion process may include assigning output values that may be indicative of an identity of the relevant spanning elements of the iteration. See, for example,FIG. 1C.

The applying, by spanning elements of an iteration that differs from the first iteration, the dimension expansion process may include assigning output values that may be indicative a history of dimension expansion processes until the iteration that differs from the first iteration.

The each spanning element may be associated with a subset of reference identifiers. The determining of the relevancy of each spanning elements of the iteration may be based a relationship between the subset of the reference identifiers of the spanning element and an output of a last merge operation before the iteration.

The output of a dimension expansion process of an iteration may be a multidimensional representation of the media unit that may include media unit regions of interest that may be associated with one or more expansion processes that generated the regions of interest.

The merge operation of the iteration may include selecting a subgroup of media unit regions of interest based on a spatial relationship between the subgroup of multidimensional regions of interest. See, for example,FIGS. 3 and 6.

Method

7000 may include applying a merge function on the subgroup of multidimensional regions of interest. See, for example,FIGS. 1C and 1F.

Method

7000 may include applying an intersection function on the subgroup of multidimensional regions of interest. See, for example,FIGS. 1C and 1F.

The merge operation of the iteration may be based on an actual size of one or more multidimensional regions of interest.

The merge operation of the iteration may be based on relationship between sizes of the multidimensional regions of interest. For example—larger multidimensional regions of interest may be maintained while smaller multidimensional regions of interest may be ignored of.

The merge operation of the iteration may be based on changes of the media unit regions of interest during at least the iteration and one or more previous iteration.

Step

7012 may be followed bystep7014 of determining identifiers that are associated with significant portions of an output of the multiple iterations.

Step

7014 may be followed bystep7016 of providing a signature that comprises the identifiers and represents the multiple objects.

Localization and Segmentation

Any of the mentioned above signature generation method provides a signature that does not explicitly includes accurate shape information. This adds to the robustness of the signature to shape related inaccuracies or to other shape related parameters.

The signature includes identifiers for identifying media regions of interest.

Each media region of interest may represent an object (for example a vehicle, a pedestrian, a road element, a human made structure, wearables, shoes, a natural element such as a tree, the sky, the sun, and the like) or a part of an object (for example—in the case of the pedestrian—a neck, a head, an arm, a leg, a thigh, a hip, a foot, an upper arm, a forearm, a wrist, and a hand). It should be noted that for object detection purposes a part of an object may be regarded as an object.

The exact shape of the object may be of interest.

FIG. 1J illustratesmethod7002 of generating a hybrid representation of a media unit.

Method

7002 may include a sequence of

steps

7020,7022,7024 and7026.

Step

7020 may include receiving or generating the media unit.

Step

7022 may include processing the media unit by performing multiple iterations, wherein at least some of the multiple iterations comprises applying, by spanning elements of the iteration, dimension expansion process that are followed by a merge operation.

Step

7024 may include selecting, based on an output of the multiple iterations, media unit regions of interest that contributed to the output of the multiple iterations.

Step

7026 may include providing a hybrid representation, wherein the hybrid representation may include (a) shape information regarding shapes of the media unit regions of interest, and (b) a media unit signature that includes identifiers that identify the media unit regions of interest.

Step

7024 may include selecting the media regions of interest per segment out of multiple segments of the media unit. See, for example,FIG. 2.

Step

7026 may includestep7027 of generating the shape information.

The shape information may include polygons that represent shapes that substantially bound the media unit regions of interest. These polygons may be of a high degree.

In order to save storage space, the method may includestep7028 of compressing the shape information of the media unit to provide compressed shape information of the media unit.

FIG. 1K illustratesmethod5002 for generating a hybrid representation of a media unit.

Method

5002 may start bystep5011 of receiving or generating a media unit.

Step

5011 may be followed by processing the media unit by performing multiple iterations, wherein at least some of the multiple iterations comprises applying, by spanning elements of the iteration, dimension expansion process that are followed by a merge operation.

The processing may be followed by steps S060 and S062.

The processing may include

steps

5020,5030,5040 and5050.

Step

5020 may include performing a k′th iteration expansion process (k may be a variable that is used to track the number of iterations).

Step

5030 may include performing a k′th iteration merge process.

Step

5040 may include changing the value of k.

Step

5050 may include checking if all required iterations were done—if so proceeding to steps S060 and S062. Else—jumping to step5020.

The output ofstep5020 is a k′th iteration expansion result.

The output ofstep5030 is a k′th iteration merge result.

Step

5060 may include completing the generation of the signature.

Step

5062 may include generating shape information regarding shapes of media unit regions of interest. The signature and the shape information provide a hybrid representation of the media unit.

The combination of steps S060 and S062 amounts to a providing a hybrid representation, wherein the hybrid representation may include (a) shape information regarding shapes of the media unit regions of interest, and (b) a media unit signature that includes identifiers that identify the media unit regions of interest.

FIG. 1L illustratesmethod5203 for generating a hybrid representation of an image.

Method

5200 may include the following sequence of steps:

Step

5210 of receiving or generating an image.

Step

5230 of performing a first iteration expansion operation (which is an expansion operation that is executed during a first iteration)

Step

5240 of performing a first iteration merge operation.

Step

5240 of amending index k (k is an iteration counter). InFIG. 1L in incremented by one—this is only an example of how the number of iterations are tracked.

Step

5280 of changing the value of index k.

Step

If yes—jumping to step5060.

Step

5060 may include completing the generation of the signature. This may include, for example, selecting significant attributes, determining retrieval information (for example indexes) that point to the selected significant attributes.

Step

5220 may include:

Step

5222 of generating multiple representations of the image within a multi-dimensional space of f(k) dimensions.

Step

5224 of assigning a unique index for each region of interest within the multiple representations. (for example, referring toFIG. 1F—indexes1 andindexes2 following first

iteration expansion operations

6031 and6032.

Step

5230 may include

Step

5226 of searching for relationships between regions of interest and define regions of interest that are related to the relationships. For example—union or intersection illustrated inFIG. 1F.

Step

5228 of assigning a unique index for each region of interest within the multiple representations. For example—referring toFIG. 1F—

indexes

1,2,12 and21.

Step

5260 may include:

Step

5270 may include

Step

FIG. 1M illustratesmethod5205 for generating a hybrid representation of an image.

Method

5200 may include the following sequence of steps:

Step

5210 of receiving or generating an image.

Step

5240 of performing a first iteration merge operation.

Step

5240 of amending index k (k is an iteration counter). InFIG. 1M in incremented by one—this is only an example of how the number of iterations are tracked.

Step

5280 of changing the value of index k.

Step

If yes—jumping to steps S060 and S062.

Step

5220 may include:

- Step5221 of filtering the image with multiple filters that are orthogonal to each other to provide multiple filtered images that are representations of the image in a multi-dimensional space of f(1) dimensions. The expansion operation ofstep5220 generates a first iteration multidimensional representation of the first image. The number of filters is denoted f(1).
- Step5224 of assigning a unique index for each region of interest within the multiple representations. (for example, referring toFIG. 1F—indexes1 andindexes2 following firstiteration expansion operations6031 and6032.

Step

5230 may include

- Step5226 of searching for relationships between regions of interest and define regions of interest that are related to the relationships. For example—union or intersection illustrated inFIG. 1F.
- Step5228 of assigning a unique index for each region of interest within the multiple representations. For example—referring toFIG. 1F—indexes1,2,12 and21.

Step

5260 may include:

- Step5262 of generating multiple representations of the merge results of the (k−1)′th iteration within a multi-dimensional space of f(k) dimensions. The expansion operation ofstep5260 generates a k′th iteration multidimensional representation of the first image. The number of dimensions of this kth iteration multidimensional representation is denoted f(k).
- Step5264 of assigning a unique index for each region of interest within the multiple representations.

Step

5270 may include

- Step5272 of searching for relationships between regions of interest and define regions of interest that are related to the relationships.
- Step5274 of assigning a unique index for each region of interest within the multiple representations.

The filters may be orthogonal may be non-orthogonal—for example be decorrelated. Using non-orthogonal filters may increase the number of filters—but this increment may be tolerable.

Object Detection Using Compressed Shape Information

Object detection may include comparing a signature of an input image to signatures of one or more cluster structures in order to find one or more cluster structures that include one or more matching signatures that match the signature of the input image.

The number of input images that are compared to the cluster structures may well exceed the number of signatures of the cluster structures. For example—thousands, tens of thousands, hundreds of thousands (and even more) of input signature may be compared to much less cluster structure signatures. The ratio between the number of input images to the aggregate number of signatures of all the cluster structures may exceed ten, one hundred, one thousand, and the like.

In order to save computational resources, the shape information of the input images may be compressed.

On the other hand—the shape information of signatures that belong to the cluster structures may be uncompressed—and of higher accuracy than those of the compressed shape information.

When the higher quality is not required—the shape information of the cluster signature may also be compressed.

Compression of the shape information of cluster signatures may be based on a priority of the cluster signature, a popularity of matches to the cluster signatures, and the like.

The shape information related to an input image that matches one or more of the cluster structures may be calculated based on shape information related to matching signatures.

For example—a shape information regarding a certain identifier within the signature of the input image may be determined based on shape information related to the certain identifiers within the matching signatures.

Any operation on the shape information related to the certain identifiers within the matching signatures may be applied in order to determine the (higher accuracy) shape information of a region of interest of the input image identified by the certain identifier.

For example—the shapes may be virtually overlaid on each other and the population per pixel may define the shape.

For example—only pixels that appear in at least a majority of the overlaid shaped should be regarded as belonging to the region of interest.

Other operations may include smoothing the overlaid shapes, selecting pixels that appear in all overlaid shapes.

The compressed shape information may be ignored of or be considered.

FIG. 1N illustratesmethod7003 of determining shape information of a region of interest of a media unit.

Method

7003 may include a sequence of

steps

7030,7032 and7034.

Step

7030 may include receiving or generating a hybrid representation of a media unit. The hybrid representation includes compressed shape information.

Step

7032 may include comparing the media unit signature of the media unit to signatures of multiple concept structures to find a matching concept structure that has at least one matching signature that matches to the media unit signature.

Step

7034 may include calculating higher accuracy shape information that is related to regions of interest of the media unit, wherein the higher accuracy shape information is of higher accuracy than the compressed shape information of the media unit, wherein the calculating is based on shape information associated with at least some of the matching signatures.

Step

7034 may include at least one out of:

Determining shapes of the media unit regions of interest using the higher accuracy shape information.

For each media unit region of interest, virtually overlaying shapes of corresponding media units of interest of at least some of the matching signatures.

FIG. 1O illustrates a matching process and a generation of a higher accuracy shape information.

It is assumed that there are multiple (M) cluster structures4974(1)-4974(M). Each cluster structure includes cluster signatures, metadata regarding the cluster signatures, and shape information regarding the regions of interest identified by identifiers of the cluster signatures.

For example—first cluster structure4974(1) includes multiple (N1) signatures (referred to as cluster signatures CS) CS(1,1)—CS(1,N1)4975(1,1)-4975(1,N1), metadata4976(1), and shape information (Shapeinfo4977(1)) regarding shapes of regions of interest associated with identifiers of the CSs.

Yet for another example—M′th cluster structure4974(M) includes multiple (N2) signatures (referred to as cluster signatures CS) CS(M,1)—CS(M,N2)4975(M,1)-4975(M,N2), metadata4976(M), and shape information (Shapeinfo4977(M)) regarding shapes of regions of interest associated with identifiers of the CSs.

The number of signatures per concept structure may change over time—for example due to cluster reduction attempts during which a CS is removed from the structure to provide a reduced cluster structure, the reduced structure is checked to determine that the reduced cluster signature may still identify objects that were associated with the (non-reduced) cluster signature—and if so the signature may be reduced from the cluster signature.

The signatures of each cluster structures are associated to each other, wherein the association may be based on similarity of signatures and/or based on association between metadata of the signatures.

Assuming that each cluster structure is associated with a unique object—then objects of a media unit may be identified by finding cluster structures that are associated with said objects. The finding of the matching cluster structures may include comparing a signature of the media unit to signatures of the cluster structures- and searching for one or more matching signature out of the cluster signatures.

InFIG. 1O—a media unit having a hybrid representation undergoes object detection. The hybrid representation includesmedia unit signature4972 andcompressed shape information4973.

Themedia unit signature4972 is compared to the signatures of the M cluster structures—from CS(1,1)4975(1,1) till CS(M,N2)4975(M,N2).

We assume that one or more cluster structures are matching cluster structures.

Once the matching cluster structures are found the method proceeds by generating shape information that is of higher accuracy then the compressed shape information.

The generation of the shape information is done per identifier.

For each j that ranges between 1 and J (J is the number of identifiers per the media unit signature4972) the method may perform the steps of:

Find (step4978(j)) the shape information of the j′th identifier of each matching signature- or of each signature of the matching cluster structure.

Generate (step4979(j)) a higher accuracy shape information of the j′th identifier.

For example—assuming that the matching signatures include CS(1,1)2975(1,1), CS(2,5)2975(2,5), CS(7,3)2975(7,3) and CS(15,2)2975(15,2), and that the j′th identifier is included in CS(1,1)2975(1,1),CS(7,3)2975(7,3) and CS(15,2)2975(15,2)—then the shape information of the j′th identifier of the media unit is determined based on the shape information associated with CS(1,1)2975(1,1),CS(7,3)2975(7,3) and CS(15,2)2975(15,2).

FIG. 1P illustrates animage8000 that includes four regions of

interest

8001,8002,8003 and8004. Thesignature8010 ofimage8000 includes variousidentifiers including ID18011,ID28012,ID38013 andID48014 that identify the four regions of

interest

8001,8002,8003 and8004.

The shapes of the four regions of

interest

8001,8002,8003 and8004 are four polygons. Accurate shape information regarding the shapes of these regions of interest may be generated during the generation ofsignature8010.

FIG. 1Q illustrates the compressing of the shape information to represent a compressed shape information that reflects simpler approximations (8001′,8002′,8003′ and8004′) of the regions of

interest

8001,8002,8003 and8004. In this example simpler may include less facets, fewer values of angles, and the like.

The hybrid representation of the media unit, after compression represent a media unit with simplified regions ofinterest8001′,8002′,8003′ and8004′—as shown inFIG. 1R.

Scale Based Bootstrap

Objects may appear in an image at different scales. Scale invariant object detection may improve the reliability and repeatability of the object detection and may also use fewer number of cluster structures—thus reduced memory resources and also lower computational resources required to maintain fewer cluster structures.

FIG. 1S illustratesmethod8020 for scale invariant object detection.

Method

8020 may include a first sequence of steps that may include

step

8022,8024,8026 and8028.

Step

8022 may include receiving or generating a first image in which an object appears in a first scale and a second image in which the object appears in a second scale that differs from the first scale.

Step

8024 may include generating a first image signature and a second image signature.

The first image signature includes a first group of at least one certain first image identifier that identifies at least a part of the object. See, forexample image8000′ ofFIG. 2A. The person is identified byidentifiers ID68016 andID88018 that represent regions of

interest

8006 and8008.

The second image signature includes a second group of certain second image identifiers that identify different parts of the object.

See, forexample image8000 ofFIG. 19. The person is identified byidentifiers ID18011,ID28012,ID38013, andID48014 that represent regions of

interest

8001,8002,8003 and8004.

The second group is larger than first group—as the second group has more members than the first group.

Step

8026 may include linking between the at least one certain first image identifier and the certain second image identifiers.

Step

8026 may include linking between the first image signature, the second image signature and the object.

Step

8026 may include adding the first signature and the second signature to a certain concept structure that is associated with the object. For example, referring toFIG. 1O, the signatures of the first and second images may be included in a cluster concept out of4974(1)-4974(M).

Step

8028 may include determining whether an input image includes the object based, at least in part, on the linking. The input image differs from the first and second images.

The determining may include determining that the input image includes the object when a signature of the input image includes the at least one certain first image identifier or the certain second image identifiers.

The determining may include determining that the input image includes the object when the signature of the input image includes only a part of the at least one certain first image identifier or only a part of the certain second image identifiers.

The linking may be performed for more than two images in which the object appears in more than two scales.

For example, seeFIG. 2B in which a person appears at three different scales—at three different images.

Infirst image8051 the person is included in a single region ofinterest8061 and thesignature8051′ offirst image8051 includes an identifier ID61 that identifies the single region of interest—identifies the person.

Insecond image8052 the upper part of the person is included in region ofinterest8068, the lower part of the person is included in region ofinterest8069 and thesignature8052′ ofsecond image8052 includes identifiers ID68 and ID69 that identify regions of

interest

8068 and8069 respectively.

Inthird image8053 the eyes of the person are included in region ofinterest8062, the mouth of the person is included in region ofinterest8063, the head of the person appears in region ofinterest8064, the neck and arms of the person appear in region ofinterest8065, the middle part of the person appears in region ofinterest8066, and the lower part of the person appears in region ofinterest8067.Signature8053′ ofthird image8053 includes identifiers ID62, ID63, ID64, ID65, ID55 and ID67 that identify regions of interest8062-8067 respectively.

Method

8020 may linksignatures8051′,8052′ and8053′ to each other. For example—these signatures may be included in the same cluster structure.

Method

8020 may link (i) ID61, (ii) signatures ID68 and ID69, and (ii) signature ID62, ID63, ID64, ID65, ID66 and ID67.

FIG. 2C illustratesmethod8030 for object detection.

Method

8030 may include the steps ofmethod8020 or may be preceded by

steps

8022,8024 and8026.

Method

8030 may include a sequence of

steps

8032,8034,8036 and8038.

Step

8032 may include receiving or generating an input image.

Step

8034 may include generating a signature of the input image.

Step

8036 may include comparing the signature of the input image to signatures of a certain concept structure. The certain concept structure may be generated bymethod8020.

Step

8038 may include determining that the input image comprises the object when at least one of the signatures of the certain concept structure matches the signature of the input image.

FIG. 2D illustratesmethod8040 for object detection.

Method

8040 may include the steps ofmethod8020 or may be preceded by

steps

8022,8024 and8026.

Method

8040 may include a sequence of

steps

8041,8043,8045,8047 and8049.

Step

8041 may include receiving or generating an input image.

Step

8043 may include generating a signature of the input image, the signature of the input image comprises only some of the certain second image identifiers; wherein the input image of the second scale.

Step

8045 may include changing a scale of the input image to the first scale to a provide an amended input image.

Step

8047 may include generating a signature of the amended input image.

Step

8049 may include verifying that the input image comprises the object when the signature of the amended input image comprises the at least one certain first image identifier.

FIG. 2E illustratesmethod8050 for object detection.

Method

8050 may include the steps ofmethod8020 or may be preceded by

steps

8022,8024 and8026.

Method

8050 may include a sequence of

steps

8052,8054,8056 and8058.

Step

8052 may include receiving or generating an input image.

Step

8054 may include generating a signature of the input image.

Step

8056 may include searching in the signature of the input image for at least one of (a) the at least one certain first image identifier, and (b) the certain second image identifiers.

Step

8058 may include determining that the input image comprises the object when the signature of the input image comprises the at least one of (a) the at least one certain first image identifier, and (b) the certain second image identifiers.

It should be noted thatstep8056 may include searching in the signature of the input image for at least one of (a) one or more certain first image identifier of the at least one certain first image identifier, and (b) at least one certain second image identifier of the certain second image identifiers.

It should be noted thatstep8058 may include determining that the input image includes the object when the signature of the input image comprises the at least one of (a) one or more certain first image identifier of the at least one certain first image identifier, and (b) the at least one certain second image identifier.

Movement Based Bootstrapping

A single object may include multiple parts that are identified by different identifiers of a signature of the image. In cases such as unsupervised learning, it may be beneficial to link the multiple object parts to each other without receiving prior knowledge regarding their inclusion in the object.

Additionally or alternatively, the linking can be done in order to verify a previous linking between the multiple object parts.

FIG. 2F illustratesmethod8070 for object detection.

Method

8070 is for movement based object detection.

Method

8070 may include a sequence of

steps

8071,8073,8075,8077,8078 and8079.

Step

8071 may include receiving or generating a video stream that includes a sequence of images.

Step

8073 may include generating image signatures of the images. Each image is associated with an image signature that comprises identifiers. Each identifier identifiers a region of interest within the image.

Step

8075 may include generating movement information indicative of movements of the regions of interest within consecutive images of the sequence of images.Step8075 may be preceded by or may include generating or receiving location information indicative of a location of each region of interest within each image. The generating of the movement information is based on the location information.

Step

8077 may include searching, based on the movement information, for a first group of regions of interest that follow a first movement. Different first regions of interest are associated with different parts of an object.

Step

8078 may include linking between first identifiers that identify the first group of regions of interest.

Step

8079 may include linking between first image signatures that include the first linked identifiers.

Step

8079 may include adding the first image signatures to a first concept structure, the first concept structure is associated with the first image.

Step

8079 may be followed by determining whether an input image includes the object based, at least in part, on the linking

An example of various steps ofmethod8070 is illustrated inFIG. 2H.

FIG. 2G illustrates three

images

8091,8092 and8093 that were taken at different points in time.

First image

8091 illustrates agate8089′ that is located in region ofinterest8089 and a person that faces the gate. Various parts of the person are located within regions of

interest

8081,8082,8083,8084 and8085.

Thefirst image signature8091′ includes identifiers ID81, ID82, ID83, ID84, ID85 and ID89 that identify regions of

interest

8081,8082,8083,8084,8085 and8089 respectively.

The firstimage location information8091″ includes the locations L81, L82, L83, L84, L85 and L89 of regions of

interest

8081,8082,8083,8084,8085 and8089 respectively. A location of a region of interest may include a location of the center of the region of interest, the location of the borders of the region of interest or any location information that may define the location of the region of interest or a part of the region of interest.

Second image

8092 illustrates a gate that is located in region ofinterest8089 and a person that faces the gate. Various parts of the person are located within regions of

interest

8081,8082,8083,8084 and8085. Second image also includes a pole that is located within region ofinterest8086. In the first image that the pole was concealed by the person.

Thesecond image signature8092′ includes identifiers ID81, ID82, ID83, ID84, ID85, ID86, and ID89 that identify regions of

interest

8081,8082,8083,8084,8085,8086 and8089 respectively.

The secondimage location information8092″ includes the locations L81, L82, L83, L84, L85, L86 and L89 of regions of

interest

8081,8082,8083,8084,8085,8086 and8089 respectively.

Third image

8093 illustrates a gate that is located in region ofinterest8089 and a person that faces the gate. Various parts of the person are located within regions of

interest

8081,8082,8083,8084 and8085. Third image also includes a pole that is located within region ofinterest8086, and a balloon that is located within region ofinterest8087.

Thethird image signature8093′ includes identifiers ID81, ID82, ID83, ID84, ID85, ID86, ID87 and ID89 that identify regions of

interest

8081,8082,8083,8084,8085,8086,8087 and8089 respectively.

The thirdimage location information8093″ includes the locations L81, L82, L83, L84, L85, L86, L87 and L89 of regions of

interest

8081,8082,8083,8084,8085,8086,8086 and8089 respectively.

The motion of the various regions of interest may be calculated by comparing the location information related to different images. The movement information may consider the different in the acquisition time of the images.

The comparison shows that regions of

interest

8081,8082,8083,8084,8085 move together and thus they should be linked to each other—and it may be assumed that they all belong to the same object.

FIG. 2H illustratesmethod8100 for object detection.

Method

8100 may include the steps ofmethod8070 or may be preceded by

steps

8071,8073,8075,8077 and8078.

Method

8100 may include the following sequence of steps:

Step

8102 of receiving or generating an input image.

Step

8104 of generating a signature of the input image.

Step

8106 of comparing the signature of the input image to signatures of a first concept structure. The first concept structure includes first identifiers that were linked to each other based on movements of first regions of interest that are identified by the first identifiers.

Step

8108 of determining that the input image includes a first object when at least one of the signatures of the first concept structure matches th e signature of the input image.

FIG. 2I illustrates method8110 for object detection.

Method8110 may include the steps ofmethod8070 or may be preceded by

steps

8071,8073,8075,8077 and8078.

Method8110 may include the following sequence of steps:

Step

8112 of receiving or generating an input image.

Step

8114 of generating a signature of the input image.

Step

8116 of searching in the signature of the input image for at least one of the first identifiers.

Step

8118 of determining that the input image comprises the object when the signature of the input image comprises at least one of the first identifiers.

Object Detection that is Robust to Angle of Acquisition

Object detection may benefit from being robust to the angle of acquisition—to the angle between the optical axis of an image sensor and a certain part of the object. This allows the detection process to be more reliable, use fewer different clusters (may not require multiple clusters for identifying the same object from different images).

FIG. 2J illustratesmethod8120 that includes the following steps:

Step

8122 of receiving or generating images of objects taken from different angles.

Step

8124 of finding images of objects taken from different angles that are close to each other. Close enough may be less than 1, 5, 10, 15 and 20 degrees—but the closeness may be better reflected by the reception of substantially the same signature.

Step

8126 of linking between the images of similar signatures. This may include searching for local similarities. The similarities are local in the sense that they are calculated per a subset of signatures. For example—assuming that the similarity is determined per two images—then a first signature may be linked to a second signature that is similar to the first image. A third signature may be linked to the second image based on the similarity between the second and third signatures- and even regardless of the relationship between the first and third signatures.

Step

8126 may include generating a concept data structure that includes the similar signatures.

This so-called local or sliding window approach, in addition to the acquisition of enough images (that will statistically provide a large angular coverage) will enable to generate a concept structure that include signatures of an object taken at multiple directions.

FIG. 2K illustrates aperson8130 that is imaged from different angles (8131,8132,8133,8134,8135 and8136). While the signature of a front view of the person (obtained from angle8131) differs from the signature of the side view of the person (obtained from angle8136), the signature of images taken from multiple angles betweenangles8141 and8136 compensates for the difference—as the difference between images obtained from close angles are similar (local similarity) to each other.

Signature Tailored Matching Threshold

Object detection may be implemented by (a) receiving or generating concept structures that include signatures of media units and related metadata, (b) receiving a new media unit, generating a new media unit signature, and (c) comparing the new media unit signature to the concept signatures of the concept structures.

The comparison may include comparing new media unit signature identifiers (identifiers of objects that appear in the new media unit) to concept signature identifiers and determining, based on a signature matching criteria whether the new media unit signature matches a concept signature. If such a match is found, then the new media unit is regarded as including the object associated with that concept structure.

It was found that by applying an adjustable signature matching criteria, the matching process may be highly effective and may adapt itself to the statistics of appearance of identifiers in different scenarios. For example—a match may be obtained when a relatively rear but highly distinguishing identifier appears in the new media unit signature and in a cluster signature, but a mismatch may be declared when multiple common and slightly distinguishing identifiers appear in the new media unit signature and in a cluster signature.

FIG. 2L illustratesmethod8200 for object detection.

Method

8200 may include:

Step

8210 of receiving an input image.

Step

8212 of generating a signature of the input image.

Step

8214 of comparing the signature of the input image to signatures of a concept structure.

Step

8216 of determining whether the signature of the input image matches any of the signatures of the concept structure based on signature matching criteria, wherein each signature of the concept structure is associated within a signature matching criterion that is determined based on an object detection parameter of the signature.

Step8218 of concluding that the input image comprises an object associated with the concept structure based on an outcome of the determining.

The signature matching criteria may be a minimal number of matching identifiers that indicate of a match. For example—assuming a signature that include few tens of identifiers, the minimal number may vary between a single identifier to all of the identifiers of the signature.

It should be noted that an input image may include multiple objects and that a signature of the input image may match multiple cluster structures.Method8200 is applicable to all of the matching processes- and that the signature matching criteria may be set for each signature of each cluster structure.

Step

8210 may be preceded bystep8202 of determining each signature matching criterion by evaluating object detection capabilities of the signature under different signature matching criteria.

Step

8202 may include:

Step

8203 of receiving or generating signatures of a group of test images.

Step

8204 of calculating the object detection capability of the signature, for each signature matching criterion of the different signature matching criteria.

Step

8206 of selecting the signature matching criterion based on the object detection capabilities of the signature under the different signature matching criteria.

The object detection capability may reflect a percent of signatures of the group of test images that match the signature.

The selecting of the signature matching criterion comprises selecting the signature matching criterion that once applied results in a percent of signatures of the group of test images that match the signature that is closets to a predefined desired percent of signatures of the group of test images that match the signature.

The object detection capability may reflect a significant change in the percent of signatures of the group of test images that match the signature. For example—assuming, that the signature matching criteria is a minimal number of matching identifiers and that changing the value of the minimal numbers may change the percentage of matching test images. A substantial change in the percentage (for example a change of more than 10, 20, 30, 40 percent) may be indicative of the desired value. The desired value may be set before the substantial change, proximate to the substantial change, and the like.

For example, referring toFIG. 1O, cluster signatures CS(1,1), CS(2,5), CS(7,3) and CS(15,2)match unit signature4972. Each of these matches may apply a unique signature matching criterion.

FIG. 2M illustratesmethod8220 for object detection.

Method

8220 is for managing a concept structure.

Method

8220 may include:

Step

8222 of determining to add a new signature to the concept structure. The concept structure may already include at least one old signature. The new signature includes identifiers that identify at least parts of objects.

Step

8224 of determining a new signature matching criterion that is based on one or more of the identifiers of the new signature. The new signature matching criterion determines when another signature matches the new signature. The determining of the new signature matching criterion may include evaluating object detection capabilities of the signature under different signature matching criteria.

Step

8224 may include

steps

8203,8204 and8206 (include din step8206) ofmethod8200.

Examples of Systems

FIG. 22N illustrates an example of a system capable of executing one or more of the mentioned above methods.

The system include various components, elements and/or units.

A component element and/or unit may be a processing circuitry may be implemented as a central processing unit (CPU), and/or one or more other integrated circuits such as application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), full-custom integrated circuits, etc., or a combination of such integrated circuits.

Alternatively, each component element and/or unit may be implemented in hardware, firmware, or software that may be executed by a processing circuitry.

System

4900 may includesensing unit4902, communication unit4904,input4911,processor4950, andoutput4919. The communication unit4904 may include the input and/or the output.

Input and/or output may be any suitable communications component such as a network interface card, universal serial bus (USB) port, disk reader, modem or transceiver that may be operative to use protocols such as are known in the art to communicate either directly, or indirectly, with other elements of the system.

Processor

4950 may include at least some out of

Multiple spanning elements4951(q).

Multiple merge elements4952(r).

4953.

4954.

4955.

4956.

Objectdetection determination unit4957.

Signature generator

4958.

Movement information unit

4959.

Identifier unit

4960.

Obstacle Avoidance

It will be appreciated that in “normal”, non-assisted driving scenarios, a driver may encounter driving obstacles that are not necessarily indicated in commonly available roadmaps. For example, it is not uncommon for a roadway to have potholes that may complicate the driving process. Some potholes are large enough to cause damage to the wheels or undercarriage of the vehicle. Accordingly, when encountering a pothole, the driver may swerve to avoid it. However, depending on the layout of the roadway and/or traffic conditions, it may not be feasible for the driver to react in such a manner. It is also possible that depending on current visibility, the speed of the vehicle, and/or the alertness of the driver, the driver may not even see the pothole in time to avoid it.

It will be appreciated that such issues are also relevant in an assisted/autonomous driving scenario. As discussed hereinabove, such systems may use one or more sensors to acquire information about the current driving environment in order to determine how to drive along the roadway as per a stored map. It is possible, that the vehicle's sensor(s) may not detect an obstacle, e.g., a pothole, sufficiently in advance in order to enable the system to determine that evasive action should be taken to avoid the obstacle and to perform that action in time. It will be appreciated that such a scenario may be further complicated by the presence of other vehicles and/or pedestrians in the immediate vicinity of the vehicle.

Reference is now made toFIG. 3A, which is a partly-pictorial, partly-block diagram illustration of an exemplary obstacle detection and mapping system10 (hereinafter referred to also as “system10”) constructed and operative in accordance with embodiments described herein. As described herein,system10 may be operative to process sensor data to detect and map obstacles on a roadway in order to enable an assisted/autonomous driving system to predict and avoid obstacles in the path of a vehicle.

System

10 comprisesvehicle100 andobstacle avoidance server400 which may be configured to communicate with each other over a communications network such as, for example, the Internet. In accordance with the exemplary embodiment ofFIG. 3A,vehicle100 may be configured with an autonomous driving system (not shown) operative to autonomously provide driving instructions tovehicle100 without the intervention of a human driver. It will be appreciated that the embodiments described herein may also support the configuration ofvehicle100 with an assisted (or “semi-autonomous”) driving system where in at least some situations a human driver may take control ofvehicle100 and/or where in at least some situations the semi-autonomous driving system provides warnings to the driver without necessarily directly controllingvehicle100.

In accordance with the exemplary embodiment ofFIG. 3A,vehicle100 may be configured with at least onesensor130 to provide information about a current driving environment asvehicle100 proceeds alongroadway20. It will be appreciated that whilesensor130 is depicted inFIG. 3A as a single entity, in practice, as will be described hereinbelow, there may bemultiple sensors130 arrayed on, or inside of,vehicle130. In accordance with embodiments described herein, sensor(s)130 may be implemented using a conventional camera operative to capture images ofroadway20 and objects in its immediate vicinity. It will be appreciated thatsensor130 may be implemented using any suitable imaging technology instead of, or in addition to, a conventional camera. For example,sensor130 may also be operative to use infrared, radar imagery, ultrasound, electro-optics, radiography, LIDAR (light detection and ranging), etc. Furthermore, in accordance with some embodiments, one ormore sensors130 may also be installed independently alongroadway20, where information fromsuch sensors130 may be provided tovehicle100 and/orobstacle avoidance server400 as a service.

In accordance with the exemplary embodiment ofFIG. 3A,static reference points30A and30B (collectively referred to hereinafter as static reference points30) may be located alongroadway20. For example,static reference point30A is depicted as a speed limit sign, and static reference point30B is depicted as an exit sign. In operation,sensor130 may capture images of static reference points30. The images may then be processed by the autonomous driving system invehicle100 to provide information about the current driving environment forvehicle100, e.g., the speed limit or the location of an upcoming exit.

Obstacle

40, e.g., a pothole, may be located onroadway20. In accordance with embodiments described herein,sensor130 may also be operative to capture images ofobstacle40. The autonomous driving system may be operative to detect the presence ofobstacle40 in the images provided bysensor130 and to determine an appropriate response, e.g., whether or notvehicle100 should change speed and/or direction to avoid or minimize the impact withobstacle40. For example, if the autonomous driving system determines thatobstacle40 is a piece of paper onroadway20, no further action may be necessary. However, ifobstacle40 is a pothole, the autonomous driving system may instructvehicle100 to slow down and/or swerve to avoidobstacle40.

It will be appreciated that there may not be sufficient processing time for the autonomous driving system to determine an appropriate response; depending on the speed ofvehicle100 and the distance at whichsensor130 captures an image ofobstacle40,vehicle100 may drive over/through obstacle40 (or collide with it) beforeobstacle40 is detected and/or an appropriate response may be determined by the autonomous driving system. In accordance with embodiments described herein, in order to provide additional processing time for obstacle detection and response determination,vehicle100 may exchange obstacle information withobstacle avoidance server400. The autonomous driving system may be operative to upload the imagery received fromsensor130 and/or the determination of the nature ofobstacle40 toobstacle avoidance server400, andobstacle avoidance server400 may be operative to provide obstacle information received in such manner fromother vehicles100 in order to enable the autonomous driving system to anticipate and respond toobstacles40 in advance of, or in parallel to, receiving the relevant imagery fromsensor130.

Depending on the configuration ofsystem10 andvehicle100,obstacle avoidance server400 andvehicle100 may exchange obstacle information in real-time (or near real-time) and/or in a burst mode, where the obstacle information may be provided either periodically and/or when conditions are suitable for a data exchange betweenobstacle avoidance server400 andvehicle100. For example,vehicle100 may be configured to perform regular uploads/downloads of obstacle information when the engine is turned on (or off, with a keep alive battery function to facilitate communication with obstacle avoidance server400). Alternatively, or in addition,vehicle100 may be configured to perform such uploads/downloads when stationary (e.g., parked, or waiting at a traffic light) and a sufficiently reliable wireless connection is detected. Alternatively, or in addition, the uploads/downloads may be triggered by location, where obstacle information may be downloaded fromobstacle avoidance server400 whenvehicle100 enters an area for which it does not have up-to-date obstacle information.

It will also be appreciated that in some situations, the autonomous driving system may not detectobstacle40 at all; it is possible thatvehicle100 may pass by/over/throughobstacle40 without detecting it. The autonomous driving system may have finite resources available to analyze imagery fromsensor130 in a finite period of time in order to determine a currently relevant driving policy. It is understandable that if, for whatever reason (e.g., driving speed, visibility, etc.), the autonomous driving system does not detectobstacle40 in real-time or near-real time there may be little benefit to be realized from continuing processing the imagery onvehicle100. However, the detection of a “missed”obstacle40 may be of benefit to anothervehicle100 onroadway20.

Accordingly, in some configurations ofsystem10,vehicle100 may also be configured to upload raw driving data fromsensor130 for processing byobstacle avoidance server400 to detectobstacles40. Obstacle information associated withobstacles40 detected in such manner may then be provided byobstacle avoidance server400 as described hereinabove.

It will also be appreciated that it may be possible to use non-image data to detectobstacles40 from the raw driving data. For example, at least onesensor130 may be implemented as an electronic control unit (ECU) controlling the shock absorbers ofvehicle100.Obstacle avoidance server400 may be configured to determine that one or more shocks of a certain magnitude and/or pattern registered by the shock absorber ECU may be indicative of a pothole.

In accordance with some embodiments described herein, the sources for raw driving data to be used byobstacle avoidance server400 to detectobstacles40 may not be limited to justvehicles100 with an autonomous driving system.Obstacle avoidance server400 may also receive raw driving data from vehicles that are being driven manually and/or with the assistance of a semi-autonomous driving system. It will be appreciated that in for such vehicles, the raw driving data may include indications of driver recognition ofobstacles40. For example, when a driver detectsobstacle40, e.g., sees a pothole, the driver may slow down in approach ofobstacle40, swerve to avoid it, and/or even change lanes. Accordingly, the raw driving data processed byobstacle avoidance server400 may also include raw data uploaded from vehicles without an autonomous driving system.

Obstacle avoidance server

400 may use supervised and/or unsupervised learning methods to detectobstacles40 in the raw driving data; reference data sets for the detection ofobstacles40 and the determination of associated driving policies may be prepared based on unsupervised analysis of raw driving data and/or using manual labelling and extraction.

It will be appreciated that it may not be a trivial task to pinpoint a location forobstacle40. While it may generally be assumed thatvehicle100 is configured with a location-based service, e.g., global positioning satellite (GPS) system, such systems may not necessarily be accurate enough to provide a sufficient resolution regarding the expected location forobstacle40. For example, one possible driving policy for evasive action may entail avoidingobstacle40 by swerving or changing lanes onroadway20. If the location-based service is only accurate to within five feet, swerving within the lane may actually increase the likelihood ofvehicle100 hitting the pothole. Similarly, it may not be possible to determine in whichlane obstacle40 is located.

In accordance with embodiments described herein, static reference points30 may be used insystem10 to more accurately determine a location forobstacle40; the location ofobstacle40 may be defined as an offset from a fixed location for one of static reference points30. For example, in the exemplary embodiment ofFIG. 3A, the location ofobstacle40 may be defined as “x” feet paststatic reference point30A, where the distance in feet may be calculated as a function of the frames per second of the camera used forsensor130 and the speed ofvehicle100. Alternatively, or in addition, twosensors130 may be employed to use triangulation to determine the distance betweenstatic reference point30A andobstacle40. Alternatively, or in addition, the distance may be calculated using bothreference points30A and30B to triangulate.

It will be appreciated that as described hereinabove,obstacle40 may be a “persistent obstacle” in that it may be reasonably expected to remain onroadway20 for an extended period of time. For example, potholes typically remain (and even increase in size) for several months before being repaired. However, some obstacles may be more transient in nature. For example, a tree fallen on the road or debris from a traffic accident may typically be removed fromroadway20 within a few hours. Such transient obstacles may, at least in some cases, be easier to detect than persistent obstacles, e.g., a fallen tree is easier to identifying from an approaching vehicle than a pothole.

It will therefore be appreciated that the driving policies for transient obstacles may be different than for persistent obstacles. For example, in a case whereobstacle40 is a pothole that may be assumed to be persistent, the associated driving policy may entail the autonomous driving system “erring on the side of caution” by performing evasive action (e.g., swerving or changing lanes) even if the pothole isn't detected in the data provided bysensor130 whenvehicle100 reaches the pothole's expected location. However, in a case whereobstacle40 is a tree onroadway20, the associated driving policy may entail the autonomous driving system performing preventative action, e.g., slowing down, prior to reaching the tree's expected location, but not to perform evasive action if the tree isn't actually detected according to the data provided bysensor130.

It will also be appreciated that some obstacles may be “recurring obstacles” where transient obstacles recur in the same location. For example, recurringobstacle50 represents a slight depression inroadway20. The impact onvehicle100 as it drives over a slight depression inroadway20 may be minimal. However, in some cases it may be observed that the slight depression may fill with water or mud when it rains, or intermittently ice over during the winter. Accordingly, the driving policies for recurringobstacles50 may more heavily weight other environmental factors, e.g., season or weather, than the driving policies forobstacles40.

Reference is now made toFIG. 3B which is a block diagram of an exemplary autonomous driving system200 (hereinafter also referred to as system200), constructed and implemented in accordance with embodiments described herein.Autonomous driving system200 comprisesprocessing circuitry210, input/output (I/O)module220,camera230,telemetry ECU240,shock sensor250,autonomous driving manager260, andobstacle database270.Autonomous driving manager260 may be instantiated in a suitable memory for storing software such as, for example, an optical storage medium, a magnetic storage medium, an electronic storage medium, and/or a combination thereof. It will be appreciated thatautonomous driving system200 may be implemented as an integrated component of an onboard computer system in a vehicle, such as, for example,vehicle100 fromFIG. 3A. Alternatively,system200 may be implemented and a separate component in communication with the onboard computer system. It will also be appreciated that in the interests of clarity, whileautonomous driving system200 may comprise additional components and/or functionality e.g., for autonomous driving ofvehicle100, such additional components and/or functionality are not depicted inFIG. 3B and/or described herein.

Processing circuitry

210 may be operative to execute instructions stored in memory (not shown). For example,processing circuitry210 may be operative to executeautonomous driving manager260. It will be appreciated that processingcircuitry210 may be implemented as a central processing unit (CPU), and/or one or more other integrated circuits such as application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), full-custom integrated circuits, etc., or a combination of such integrated circuits. It will similarly be appreciated thatautonomous driving system200 may comprise more than one instance ofprocessing circuitry210. For example, one such instance ofprocessing circuitry210 may be a special purpose processor operative to executeautonomous driving manager260 to perform some, or all, of the functionality ofautonomous driving system200 as described herein.

I/O module220 may be any suitable communications component such as a network interface card, universal serial bus (USB) port, disk reader, modem or transceiver that may be operative to use protocols such as are known in the art to communicate either directly, or indirectly, with other elements of system10 (FIG. 3A) and/orsystem200, such as, for example, obstacle avoidance server400 (FIG. 3A),camera230,telemetry ECU240, and/orshock sensor250. As such, I/O module220 may be operative to use a wired or wireless connection to connect toobstacle avoidance server400 via a communications network such as a local area network, a backbone network and/or the Internet, etc. I/O module220 may also be operative to use a wired or wireless connection to connect to other components ofsystem200, e.g.,camera230,telemetry ECU240, and/orshock sensor250. It will be appreciated that in operation I/O module220 may be implemented as a multiplicity of modules, where different modules may be operative to use different communication technologies. For example, a module providing mobile network connectivity may be used to connect toobstacle avoidance server400, whereas a local area wired connection may be used to connect tocamera230,telemetry ECU240, and/orshock sensor250.

In accordance with embodiments described herein,camera230,telemetry ECU240, andshock sensor250 represent implementations of sensor(s)130 fromFIG. 3A. It will be appreciated thatcamera230,telemetry ECU240, and/orshock sensor250 may be implemented as integrated components of vehicle100 (FIG. 3A) and may provide other functionality that is the interests of clarity is not explicitly described herein. As described hereinbelow,system200 may use information about a current driving environment as received fromcamera230,telemetry ECU240, and/orshock sensor250 to determine an appropriate driving policy forvehicle100.

Autonomous driving manager

260 may be an application implemented in hardware, firmware, or software that may be executed by processingcircuitry210 to provide driving instructions tovehicle100. For example,autonomous driving manager260 may use images received fromcamera230 and/or telemetry data received fromtelemetry ECU240 to determine an appropriate driving policy for arriving at a given destination and provide driving instructions tovehicle100 accordingly. It will be appreciated thatautonomous driving manager260 may also be operative to use other data sources when determining a driving policy, e.g., maps of potential routes, traffic congestion reports, etc.

As depicted inFIG. 3B,autonomous driving manager260 comprisesobstacle detector265,obstacle predictor262, andobstacle avoidance module268. It will be appreciated that the depiction ofobstacle detector265,obstacle predictor262, andobstacle avoidance module268 as integrated components ofautonomous driving manager260 may be exemplary. The embodiments described herein may also support implementation ofobstacle detector265,obstacle predictor262, andobstacle avoidance module268 as independent applications in communication withautonomous driving manager260, e.g., via I/O module220.

Obstacle detector

265,obstacle predictor262, andobstacle avoidance module268 may be implemented in hardware, firmware, or software and may be invoked byautonomous driving manager260 as necessary to provide input to the determination of an appropriate driving policy forvehicle100. For example,obstacle detector265 may be operative to use information from sensor(s)130 (FIG. 3A), e.g.,camera230,telemetry ECU240, and/orshock sensor250 to detect obstacles in (or near) the driving path ofvehicle100, e.g., along (or near) roadway20 (FIG. 3A).Obstacle predictor262 may be operative to use obstacle information received fromobstacle avoidance server400 to predict the location of obstacles along or nearroadway20 before, or in parallel to their detection byobstacle detector265.Obstacle avoidance module268 may be operative to determine an appropriate driving policy based at least on obstacles detected/predicted (or not detected/predicted) byobstacle detector265 and/orobstacle predictor262.

Autonomous driving manager

260 may store obstacle information received fromobstacle avoidance server400 inobstacle database270 for use byobstacle detector265,obstacle predictor262, andobstacle avoidance module268 as described herein. It will be appreciated that obstacle information, including driving policies to avoid detected obstacles may also be stored inobstacle database270 for use byobstacle detector265,obstacle predictor262, andobstacle avoidance module268.

Reference is now made also toFIG. 3C which is a flowchart of an exemplary obstacle detection and avoidance process300 (hereinafter also referred to as process300) to be performed bysystem200 during (or preceding) a driving session for vehicle100 (FIG. 3A). The components of

systems

100 and200 will be referred to herein as per the reference numerals inFIGS. 3A and 3B.

In accordance with some embodiments described herein,system200 may receive (step310) an obstacle warning fromobstacle avoidance server400, e.g. obstacle information received fromother vehicles100 usingsystem10 and/or and other suitable sources as described with reference toFIG. 3A. It will be appreciated that the obstacle warning may be received by I/O module220. Depending on the configuration ofsystem100, the obstacle warning may be received in a batch update process, either periodically and/or triggered by an event, e.g., whenvehicle100 is turned on, whenvehicle100 enters a new map area, whenvehicle100 enters an area with good wireless reception, etc. It will be appreciated that the obstacle warning received instep310 may also include a description of the obstacle, e.g., a pothole.

Autonomous driving manager

260 may invokeobstacle predictor262 to estimate (step320) a location for the obstacle associated with the obstacle warning. It will be appreciated that, as discussed hereinabove, while the obstacle warning may include GPS coordinates for the obstacle, the coordinates may not be accurate enough to provide the location with sufficient precision to be of use in an automated. Furthermore, even if the coordinates for the obstacle, e.g., obstacle40 (FIG. 3A) are accurate,vehicle10 will likely be in movement and it may not be possible to accurately determine the GPS coordinates ofvehicle10 as it approaches the obstacle.

In accordance with embodiments described herein, the obstacle information in the obstacle warning may include landmarks to be used for calculating the obstacle location. For example, the location of the associated obstacle may be indicated in the obstacle warning as an offset from a local landmark, e.g., one or more of static reference points30 (FIG. 3A). It will be appreciated that static reference points30 may be visible from a greater distance thanobstacle40; a signpost typically extends vertically from the ground such that static reference points30 may be captured by sensor(s)130, e.g.,camera230, in advance of detection ofobstacle40. Obstacle predictor may then use static reference points30 to predict a location for the associatedobstacle40 as an offset from the observed location of one or more static reference points30 before sensor(s)130 may actually provide an indication of the location. Distance from a given static reference point30 may be estimated as a function of current speed forvehicle10 and a frames-per-second setting forcamera230. Alternatively, or in addition, multiple static reference points may be used to locateobstacle40 by triangulation, e.g., a distance from bothstatic reference points30A and30B may be estimated and then used to predict the location ofobstacle40.

It will be appreciated that in some embodiments, sensor(s)130 may be operative to provide an estimate for distance. For example, sensor(s) may be operative to use radar imagery, LIDAR, etc.

Autonomous driving manager

260 may invokeobstacle avoidance module268 to enact (step330) preventative measures to avoid colliding with, or running over,obstacle40.Obstacle avoidance module268 may enact the preventative measures in accordance with driving policies stored inobstacle database270. It will be appreciated that the obstacle warning may include a description ofobstacle40, e.g., a pothole. Using the obstacle description and other available information (e.g., distance toobstacle40 as estimated instep320, current speed, road conditions, etc.),obstacle avoidance module268 may select an appropriate driving policy fromobstacle database270 for implementation. For example, the selected driving policy may entail a reduction in speed, switching lanes (if possible, under current conditions), swerving within a lane, etc. It will be appreciated that the driving policies in obstacle database may be pre-stored inobstacle database270. It will further be appreciated that the driving policies may be derived based on supervised or unsupervised analysis of imagery from actual driving sessions. Some, or all, of the driving policies may also be defined manually.

In some cases, the selected driving policy may entail more extreme preventative measures. For example,obstacle40 may be described in the obstacle warning as a very large obstacle that may effectively block traffic onroadway20, e.g., a rockfall, or debris from a multi-vehicle accident. In such a case, the driving policy may entail actions such as, for example, pulling over on the shoulder, stopping in place, or changing routes.

Autonomous driving manager may invokeobstacle detector265 to determine if an obstacle, e.g.,obstacle40, has been detected (step340). For example,obstacle detector265 may use imagery received from sensor(s)130 to detectobstacle40 inroadway20.

It will be appreciated that in typical operation ofprocess300, steps310-330 may not always be performed;autonomous driving system200 may not always receive an obstacle warning in advance of encounteringobstacle40. For example,vehicle100 may be the first vehicle associated withsystem10 to encounterobstacle40. It will therefore be appreciated thatautonomous driving manager260 may continuously executeprocess300 in an attempt to detect and avoid obstacles asvehicle100 progresses alongroadway20. Accordingly, step340 may be performed independently of whether or not an obstacle warning has been received for a given area onroadway20. It will, however, be appreciated, that in the event that an obstacle warning was indeed received, the relevant obstacle information may be used byobstacle detector265 to determine whether or not the indicated obstacle, e.g.,obstacle40, is indeed in place onroadway20 as indicated in the obstacle warning.

Alternatively, or in addition,obstacle detector265 may be detected in response to information received fromshock sensor250.Shock sensor250 may be a sensor operative to monitor the operation of the shock absorbers onvehicle100. It will be appreciated that when running over a pothole, shock absorbers may absorb the ensuing shock and stabilizevehicle100. It will also be appreciated that in some situations, potholes may not be detected untilvehicle100 actually drives over them. In such a situation,obstacle detector265 may use information fromshock sensor250 to detectobstacle40, albeit aftervehicle100 has already hitobstacle40.

Ifobstacle40 is detected instep340,autonomous driving manager260 may invokeobstacle avoidance module268 to determine whetherobstacle40 is in the path of vehicle100 (step350). It will be appreciated that someobstacles40 onroadway20 may not be in the direct path ofvehicle100; someobstacles40 may be in a different lane thanvehicle100. Additionally, ifstep330 was performed,vehicle100 may have already switched lanes in advance ofobstacle40 being detected instep340. Furthermore, ifobstacle40 was detected usingshock sensor250,vehicle100 may have already run overobstacle40 such that the result ofstep350 may be “no”. Ifobstacle40 is not detected instep340, process control may flow through to step380.

If, as per the result ofstep350,obstacle40 is in the path ofvehicle100,obstacle avoidance module268 may determine whetherobstacle40 is avoidable (step360). It will be appreciated thatobstacle40 may not be detected instep340 untilvehicle100 is very close. For example, visibility may be poor,obstacle40 may be located immediately around a bend inroadway20,obstacle40 may be effectively camouflaged by its surroundings (e.g., some potholes are not distinguishable from a distance), etc. Accordingly, by thetime obstacle40 is detected, there may not be enough time to enact avoidance measures prior to impact. If, as per the result ofstep350,obstacle40 is not avoidable, process control may flow through to step380.

If, as per the result ofstep360,obstacle40 is avoidable,obstacle avoidance module268 may instructvehicle100 to avoid (step370)obstacle40. For example,obstacle avoidance module268 may instructvehicle100 to straddleobstacle40, swerve within the lane, or switch lanes altogether.

If, as per the result ofstep360,obstacle40 is not avoidable,obstacle avoidance module268 may instructvehicle100 to minimize (step365) impact withobstacle40. For example,obstacle avoidance module268 may instructvehicle100 to reduce speed and/or to swerve to avoid hitting the center ofobstacle40. Process control may then flow through to step380.

Autonomous driving manager

380 may use I/O module220 to send an obstacle report toobstacle avoidance server400. The obstacle report may, for example, include telemetry data and sensor imagery that may be used byobstacle avoidance server400 to determine a location forobstacle40. For example, the obstacle report may include camera images (from camera230) and telemetry data (from telemetry ECU240) from a period staring a given number of seconds beforeobstacle40 was detected instep340. Alternatively, or in addition, the telemetry data and camera images may be from a period defined according to a physical distance fromobstacle40. The obstacle report may also include data fromshock sensor250 where applicable.

Depending on the configuration of

systems

100 and200,step380 may be performed in real time or near real time afterobstacle40 is detected instep340. Alternatively, or in addition,step380 may be performed in a periodic and/or event triggered batch process.

It will be appreciated that obstacle reports may also be provided for “no” results inprocess300. Ifobstacle40 is not detected instep340, an obstacle report may be provided toobstacle avoidance server400 indicating that an obstacle was not detected at the associated location. It will be appreciated that in some cases, a “no” result instep340 may indicate that an obstacle associated with an obstacle warning may no longer be found onroadway20, e.g., the pothole may have been filled in. In other cases, where an obstacle warning was not received, the “no” result may indicate that the default condition, e.g., “no obstacles”, is in effect forroadway20.

Reference is now made toFIG. 4 which is a block diagram of an exemplary obstacle avoidance server400 (hereinafter also referred to as server400), constructed and implemented in accordance with embodiments described herein.Server400 comprisesprocessing circuitry410, input/output (I/O)module420,obstacle avoidance manager460, andobstacle database470.Obstacle avoidance manager460 may be instantiated in a suitable memory for storing software such as, for example, an optical storage medium, a magnetic storage medium, an electronic storage medium, and/or a combination thereof.

Processing circuitry

410 may be operative to execute instructions stored in memory (not shown). For example,processing circuitry410 may be operative to executeobstacle avoidance manager260. It will be appreciated that processingcircuitry410 may be implemented as a central processing unit (CPU), and/or one or more other integrated circuits such as application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), full-custom integrated circuits, etc., or a combination of such integrated circuits. It will similarly be appreciated thatserver400 may comprise more than one instance ofprocessing circuitry410. For example, one such instance ofprocessing circuitry410 may be a special purpose processor operative to executeobstacle avoidance manager460 to perform some, or all, of the functionality ofserver400 as described herein.

I/O module420 may be any suitable communications component such as a network interface card, universal serial bus (USB) port, disk reader, modem or transceiver that may be operative to use protocols such as are known in the art to communicate either directly, or indirectly, with other elements of system10 (FIG. 3A) such as, for example, system200 (FIG. 3B). As such, I/O module420 may be operative to use a wired or wireless connection to connect tosystem200 via a communications network such as a local area network, a backbone network and/or the Internet, etc. It will be appreciated that in operation I/O module220 may be implemented as a multiplicity of modules, where different modules may be operative to use different communication technologies. For example, a module providing mobile network connectivity may be used to connect wirelessly to one instance ofsystem200, e.g., one vehicle100 (FIG. 3A), whereas a local area wired connection may be used to connect to a different instance ofsystem100, e.g., adifferent vehicle100.

Obstacle avoidance manager

460 may be an application implemented in hardware, firmware, or software that may be executed by processingcircuitry410 to provide obstacle warnings and/or driving policies tovehicles100. For example,obstacle avoidance manager460 may use obstacle information in obstacle reports received fromvehicles100 to provide obstacle warnings in real time, near real time, and/or batch mode tovehicles100 in system10 (FIG. 3A). It will be appreciated thatobstacle avoidance manager460 may also be operative to use other data sources when detecting obstacles and determining driving policies, e.g., maps of potential routes, traffic congestion reports, etc.

As depicted inFIG. 4,obstacle avoidance manager460 comprisesobstacle extractor462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468. It will be appreciated that the depiction ofobstacle extractor462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468 as integrated components ofobstacle avoidance manager460 may be exemplary. The embodiments described herein may also support implementation ofobstacle extractor462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468 as independent applications in communication withobstacle avoidance manager460, e.g., via I/O module420.

Obstacle extractor

462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468 may be implemented in hardware, firmware, or software and may be invoked byobstacle avoidance manager460 as necessary to provide obstacle warnings and associated driving policies tovehicles100. For example,obstacle extractor462 may be operative to extract obstacle information from obstacle reports received fromvehicles100.Obstacle categorizer464 may be operative to use categorize obstacles in terms of danger level and/or persistence characteristics.Obstacle policy manager466 may be operative to determine an appropriate driving policy for a given obstacle or category of obstacle.Obstacle timer468 may be operative to set an expiration timer for obstacles according to their associated obstacle categories.

Obstacle avoidance manager

460 may store obstacle information received fromvehicles100 inobstacle database270 for use byobstacle extractor462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468 as described herein. It will be appreciated thatobstacle database470 may also be used to store obstacle information, including driving policies, derived byobstacle extractor462,obstacle categorizer464,obstacle policy manager466, andobstacle timer468.

Reference is now made also toFIG. 5 which is a flowchart of an exemplary obstacle detection and warning process500 (hereinafter also referred to as process500) to be performed byserver400 to support driving sessions for vehicles100 (FIG. 3A). The components of

systems

10 and200 will be referred to herein as per the reference numerals inFIGS. 3A and 3B.1 and2. Similarly, the components ofserver400 will be referred to herein as per the reference numerals inFIG. 4.

In accordance with some embodiments described herein,server400 may receive (step510) an obstacle report fromvehicle100. It will be appreciated that the obstacle warning may be received by I/O module420. Depending on the configuration ofsystem10, the obstacle report may be received in real time or near real time aftervehicle100

encounters obstacle

40. Alternatively, or in addition, the obstacle report may be received in a batch update process, either periodically and/or triggered by an event, e.g., whenvehicle100 is turned on, whenvehicle100 enters a new map area, whenvehicle100 enters an area with good wireless reception, etc. It will be appreciated that the obstacle report received instep510 may also include an image ofobstacle40.

It will be appreciated that as described hereinabove with respect to process300 (FIG. 3C), a given obstacle report may not necessarily be triggered by detection of an obstacle byvehicle100. The obstacle report may represent a periodic upload of GPS data, camera images, telemetry data, and/or other sensor data that may not be associated with a specific detection of an obstacle. The obstacle report may also be associated with a non-detection of an obstacle that was otherwise indicated by a previously issued obstacle warning. Furthermore, as described with respect toFIG. 3A, raw driving data from manually and/or semi-autonomously driven vehicles may also be provided toserver400. In accordance with embodiments described herein, such raw driving data may also be input to process500 along with obstacle reports.

Accordingly,obstacle avoidance server400 may invokeobstacle extractor462 to analyze received obstacle reports and/or raw driving data to determine if a given obstacle report (or raw driving data) includes a reported obstacle (step520), i.e., where the obstacle report specifically addresses an obstacle, whether it was detected byvehicle100 and/or it was included in a previous obstacle warning.

Obstacle extractor

462 may determine if the obstacle report or raw driving data includes at least a reference to a reported obstacle (step520). It will be appreciated that as described with respect to process300, an obstacle report may be sent in response to an obstacle warning, even if the associated obstacle is not actually detected byautonomous driving manager360. Such an obstacle report may still include a reference to the reported obstacle. If the obstacle report or raw driving data includes at least a reference to a reported obstacle (step520) processing may continue withstep530.

If the obstacle report or raw driving data does not include a reference to a reported obstacle (step520),obstacle extractor462 may process (step525) the obstacle report (or raw driving data) to extract obstacle information for obstacles. For example,obstacle extractor464 may be operative to detect an obstacle based on an image of an obstacle, where the image may be matched to a reference image of an obstacle stored inobstacle database470. Alternatively, or in addition,obstacle extractor464 may be operative to detect an obstacle based on telemetry data, where certain driving patterns maybe indicative of a driver avoiding an obstacle. Alternatively, or in addition,obstacle extractor464 may be operative to detect an obstacle based on analysis of shock absorber data, where shock events may be indicative of a vehicle running over a pothole. It will be appreciated that the obstacle information extracted instep525 may also include reference data that may be used to provide a location for the extracted obstacle, e.g., GPS coordinates, offsets from fixed reference points30, radar imagery, etc.

If one or more obstacles are extracted (step529) processing control may flow through to step530. Otherwise, processing control may return to step510.

Obstacle avoidance manager

460 may determine whether a reported obstacle or an extracted obstacle is a “new” obstacle (step530), i.e., whether the associated obstacle is already known toserver400 from a previous iteration ofprocess300. If the obstacle is a new obstacle,obstacle avoidance manager460 may invokeobstacle categorizer464 to categorize the obstacle.Obstacle categorizer464 may categorize the obstacle based on a level of danger it may represent tovehicle100. For example, if the obstacle is a relatively shallow pothole, obstacle categorizer may assign it to a low-risk category. If the obstacle is a patch of ice, obstacle categorizer may assign it to a medium-risk category. If the obstacle is a large slab of concrete, obstacle categorizer may assign it to a high-risk category. It will be appreciated that the examples provided herein for obstacles and their associated categories are not limiting; the embodiments described herein may support other combinations of obstacles and categories.

Obstacle categorizer

464 may also be operative to categorize obstacles based on their expected persistence. For example, a pothole may be a persistent obstacle; it may be expected to remain onroadway20 for weeks, months, or even years after first being observed. Debris from a traffic accident may be a temporary obstacle; it may be expected to be cleared fromroadway20 within hours or perhaps a day or two after first being observed. Some obstacles may be recurring in nature; such obstacles may be observed intermittently in the same location. For example, puddles or ice patches may tend to recur in the same locations based on the topography ofroadway20. Some obstacles may have hybrid persistence characteristics. For example, a pothole may also tend to fill with rain/slow and freeze during the winter. The pothole itself may be of a persistent nature, whereas the resulting ice patch (or puddle) may be of a recurring nature.Obstacle categorizer464 may categorize (step535) the obstacle accordingly, e.g. as persistent, temporary, recurring, or hybrid.

Obstacle avoidance manager

460 may invokeobstacle policy manager466 to set (step539), based at least on the obstacle's category, an appropriate driving policy for avoiding (or at least minimizing the risks associated with) the obstacle.

If, perstep530, the obstacle is not a new obstacle,obstacle avoidance manager460 may use obstacle information stored inobstacle database470 to determine whether the current obstacle information is consistent with previously observed obstacle information (step540). For example, a pothole may expand in size and/or fill with water/ice compared to a previous observation. Debris from a traffic accident may shift or be partially removed. If the current obstacle information is inconsistent with the stored obstacle information,obstacle avoidance manager460 may invokeobstacle categorizer464 to update (step545) the category or categories assigned instep535, andobstacle policy manager466 to update (step549) the driving policy set instep539.

As described hereinabove, even though a given obstacle report may be associated with a previously issued obstacle warning, in some cases the obstacle report may not include sensor data indicative of an observation of the associated obstacle.Vehicle100 may receive an obstacle warning, but by thetime vehicle100 reaches the location indicated in the obstacle warning, the associated obstacle may no longer be there. For example, based on obstacle information received byobstacle avoidance server400,obstacle avoidance manager460 may determine that debris from traffic accident may be at a given location onroadway20, and send an obstacle warning tovehicle100 for the given location. However, the debris may be cleared by the time thatvehicle100 reaches the given location. As described with reference to process300, in such a case,vehicle100 may provide an obstacle report toserver400 even though it did not actually detect the obstacle associated with the obstacle warning.

Accordingly,obstacle avoidance manager460 may determine if an obstacle associated with the obstacle report was actually detected byautonomous driving system200 or obstacle extractor462 (step550). If an obstacle was detected perstep550,obstacle avoidance manager460 may useobstacle timer468 to set (step555) a timer for the detected obstacle. The timer may be set according to an anticipated duration for the obstacle. For example, if the obstacle is debris from a traffic accident, the timer may be set to a number of hours; for a pothole the timer may be set for a number of weeks or months.

It will be appreciated that the detected obstacle fromstep550 may already have had a timer set in a previous iteration ofprocess500, e.g., the obstacle report is associated with a known obstacle for which an obstacle warning was previously generated and sent to vehicle(s)100. In such a case, the timer may be reset to extend the anticipated duration of the obstacle per a more recent observation. For example, if the obstacle was debris from a traffic accident, the timer may have originally been set for two hours. If an obstacle report is received ninety minutes later indicating that the debris is still there,obstacle avoidance manager460 may reset the timer to update the anticipated duration based on the latest information (i.e., the recent obstacle report).

If an obstacle associated with the obstacle report was not detected byautonomous driving system200 or obstacle extractor462 (step550),obstacle avoidance manager460 may determine if the obstacle warning has expired (step560) according to the timer (re)set instep555. If the timer has expired,obstacle avoidance manager460 may remove (step465) the obstacle from a warning list of currently relevant obstacle warnings. Otherwise, processing control may continue to step570.

Obstacle avoidance manager

460 may then update (step570) the warning list as necessary e.g., adding obstacle warnings, modifying driving policies, removing obstacle warnings, etc.Obstacle avoidance manager460 may then send (step580) the obstacle warning(s) from the warning list to vehicle(s)100. Depending on the configuration of

systems

10,200, and400,step580 may be performed in real time or near real time afterstep570. Alternatively, or in addition,step580 may be performed in a periodic and/or event triggered batch process.

Process control may then return to step510. It will be appreciated that in some embodiments, steps520-570 (or combinations, thereof) may be performed iteratively, e.g., if multiple obstacles are extracted from raw driving data.

Obstacle Detection Based on Visual and Non-Visual Information

Obstacles may be viewed as elements that introduce changes in the behavior of a vehicle. Obstacles may be automatically detected by processing images obtained while a vehicle performs maneuvers that are suspected as being obstacle avoidance maneuvers.

FIG. 6 illustrates amethod1600 executed by computerized system for detecting obstacles. The computerized system may be located outside a vehicle, may be distributed between different vehicles, and the like.

Method

1600 may start bystep1610 of receiving, from a plurality of vehicles, and by an I/O module of a computerized system, visual information acquired during executions of vehicle maneuvers that are suspected as being obstacle avoidance maneuvers.

Step

1610 may also include receiving behavioral information regarding behavior of the plurality of vehicles, during an execution of the maneuvers that are suspected as being obstacle avoidance maneuvers. The behavioral information may represent the behavior (speed, direction of propagation, and the like) of the entire vehicle and/or may represent the behavior of parts or components of the vehicle (for example damping a shock by a shock absorber, slowing the vehicle by the breaks, turns of the steering wheel, and the like), and/or may represent the behavior of the driver (the manner in which the driver drives), and the like.

A determination of what constitutes a vehicle maneuver suspected as being an obstacle avoidance maneuver may be detected in a supervised manner or in an unsupervised manner. For example—when insupervised manner method1600 may include receiving examples of maneuvers that are tagged as obstacle avoidance maneuvers.

Such a maneuver may be an uncommon maneuver in comparison to at least one out of (a) maneuvers of the same vehicle, (b) maneuvers performed at the same location by other drivers, (c) maneuvers performed in other places by other drivers, and the like. The maneuver may be compared to other maneuvers at the same time of day, at other times of day, at the same date, at other dates and the like.

The vehicle maneuver suspected as being an obstacle avoidance maneuver may be detected based on behavioral information that may be obtained from one or more sensors such as an accelerometer, a wheel speed sensor, a vehicle speed sensor, a brake sensor, a steering wheel sensor, a shock absorber sensor, an engine sensor, a driver sensor for sensing one or more physiological parameter of the driver (such as heart beat), or any other sensor—especially any other non-visual sensor.

The vehicle maneuver suspected as being an obstacle avoidance maneuver may involve a change (especially a rapid change) in the direction and/or speed and/or acceleration of the vehicle, a deviation from a lane, a deviation from a previous driving pattern followed at attempt to correct the deviation, and the like.

For example—when applying supervised learning of obstacles—the computerized system may be fed with examples of vehicle behaviors when bypassing known obstacles, when approaching a significant bump, and the like.

For example—the determining may include ruling out common behaviors such as stopping in front of a red traffic light.

For example—a fast and significant change in the speed and direction of a vehicle at a linear road path may indicate that the vehicle encountered an obstacle.

The visual information may be raw image data (such as images) acquired by a visual sensor of the vehicle, processed images, metadata or any other type of visual information that represents the images acquired by visual sensor.

The visual information may include one or more robust signatures of one or more images acquired by one or more visual sensors of the vehicle. An example of a robust signature is illustrated in U.S. Pat. No. 8,326,775 which is incorporated herein by reference.

Step

1610 may be followed bystep1620 of determining, based at least on the visual information, at least one visual obstacle identifier for visually identifying at least one obstacle.

The determining may include clustering visual information based on the objects represented by the visual information, and generating a visual obstacle identifier that represents a cluster. The clusters may be linked to each other and a visual obstacle identifier may represent a set of linked clusters.

The determining may include filtering out objects that are represented in the visual information based on a frequency of appearance of the objects in the visual information. For example—the road may appear in virtually all images and may be ignored.

The clustering may or may not be responsive to the behavior of the vehicle during the executions of vehicle maneuvers that are suspected as being obstacle avoidance maneuvers.

The visual obstacle identifier may be a model of the obstacle, a robust signature of the obstacle, configuration information of a neural network related to the obstacle.

A model of the obstacle may define one or more parameters of the obstacle such as shape, size, color of pixels, grayscale of pixels, and the like.

The configuration information of a neural network may include weights to be assigned to a neural network. Different neural networks may be used to detect different obstacles.

The visual obstacle identifier for visually identifying an obstacle may identify a group of obstacles to which the obstacle belongs. For example—the visual obstacle identifier may identify a concept. See, for example, U.S. Pat. No. 8,266,185 which is incorporated herein by reference.

A visual obstacle identifier that identifies an obstacle may include at least one out of severity metadata indicative of a severity of the obstacle, type metadata indicative of a type of the obstacle (for example—pothole, tree, bump), size metadata indicative of a size of the obstacle, timing metadata indicative of a timing of an existence of the obstacle (for example—persistent, temporary, recurring), and location metadata indicative of a location of the obstacle. See, forexample severity metadata1851,type metadata1852,size metadata1853,timing metadata1854 andlocation metadata1855 ofFIG. 7.

Step

1620 may be followed bystep1630 of responding to the outcome ofstep1620. For example—step1630 may include transmitting to one or more of the plurality of vehicles, the at least one visual obstacle identifier. Additionally or alternatively,step1630 may include populating an obstacle database. The visual obstacle identifier may be included in an obstacle warning.

The behavioral information is obtained by non-visual sensors of the plurality of vehicles.

Step

1630 may also include verifying, based on at least the visual information, whether the maneuvers that are suspected as being obstacle avoidance maneuvers are actually obstacle avoidance maneuvers.

The verification may include detecting at least a predefined number of indications about a certain obstacle.

The verification may include receiving information from another source (for example input from drivers, information provided from the police, department of transportation or other entity responsible to the maintenance of roads, regarding the existence of obstacles.

Step

1630 may be followed bystep1640 of transmitting to one or more of the plurality of vehicles, verification information indicative of the maneuvers that are actually obstacle avoidance maneuvers. This may help the vehicles to ignore maneuvers that were suspected as obstacle avoidance maneuvers—which are not actual obstacle avoidance maneuvers.

Method

1600 enables a vehicle that receives a visual obstacle identifier to identify an obstacle of known parameters—even if the vehicle is the first one to image that obstacle.Method1600 may be executed without human intervention and may detect obstacles without prior knowledge of such obstacles. This provides a flexible and adaptive method for detecting obstacles. The visual information and/or the visual obstacle identifier may be very compact—thereby allowing the learning of obstacles and afterwards detecting obstacles to be executed using a relatively small amount of computational and/or storage and/or communication resources.

It should be noted that different vehicle may perform obstacle avoidance maneuvers that may differ from each other by duration, vehicle behavior and the like. An example is illustrated inFIGS. 8-10 in which first, second and third vehicles perform different obstacle avoidance maneuvers and acquire different numbers of images during said obstacle avoidance maneuvers.

FIG. 8 illustrates a first vehicle (VH1)1801 that propagates along aroad1820.First vehicle1801 performs amaneuver1832 suspected as being an obstacle avoidance maneuver when encountered withobstacle1841.Maneuver1832 is preceded by anon-suspected maneuver1831 and is followed by anothernon-suspected maneuver1833.

First vehicle

1801 acquires a first plurality (N1) of images I1(1)-I1(N1)1700(1,1)-1700(1,N1) duringobstacle avoidance maneuver1832.

Visual information V1(1)-V1(N1)1702(1,1)-1702(1,N1) is sent fromfirst vehicle1801 to computerized system (CS)1710 vianetwork1720.

The visual information may be the images themselves. Additionally or alternatively, first vehicle processes the images to provide a representation of the images.

First vehicle

1801 may also transmit behavioral information B1(1)-B1(N1)1704(1,1)-1704(1,N1) that represents the behavior of the vehicle duringmaneuver1832.

FIG. 9 illustrates a second vehicle (VH2)1802 that propagates along aroad1820.Second vehicle1802 performs amaneuver1833 suspected as being an obstacle avoidance maneuver when encountered withobstacle1841.Maneuver1832 is preceded by a non-suspected maneuver and is followed by another non-suspected maneuver.

Second vehicle

1802 acquires a second plurality (N2) of images I2(1)-I2(N2)1700(2,1)-1700(2,N2) duringmaneuver1833.

Visual information V2(1)-V2(N2)1702(2,1)-1702(2,N2) is sent fromsecond vehicle1802 to computerized system (CS)1710 vianetwork1720.

The visual information may be the images themselves. Additionally or alternatively, second vehicle processes the images to provide a representation of the images.

Second vehicle

1802 may also transmit behavioral information (not shown) that represents the behavior of the vehicle duringmaneuver1832.

FIG. 10 illustrates a third vehicle (VH2)1803 that propagates along a road.Third vehicle1803 performs amaneuver1834 suspected as being an obstacle avoidance maneuver when encountered withobstacle1841.Maneuver1834 is preceded by a non-suspected maneuver and is followed by another non-suspected maneuver.

Third vehicle

1803 acquires a third plurality (N3) of images I3(1)-I3(N3)1700(3,1)-1700(3,N3) duringmaneuver1834.

Visual information V3(1)-V3(N3)1702(3,1)-1702(3,N3) is sent fromthird vehicle1803 to computerized system (CS)1710 vianetwork1720.

The visual information may be the images themselves. Additionally or alternatively, third vehicle processes the images to provide a representation of the images.

Third vehicle

1803 may also transmit behavioral information (not shown) that represents the behavior of the vehicle duringmaneuver1832.

FIG. 11 illustrates a method1900 for detecting obstacles. The method may be executed by a vehicle.

Method1900 may start by steps1910 and1920.

Step1910 may include sensing, by a non-visual sensor of a vehicle, a behavior of a vehicle. The non-visual sensor may be an accelerometer, a shock absorber sensor, a brakes sensor, and the like.

Step1920 may include acquiring, by a visual sensor of the vehicle, images of an environment of the vehicle. Step1920 may continue even when such a suspected maneuver is not detected.

Steps1920 and1930 may be followed by step1930 of determining, by a processing circuitry of the vehicle, whether the behavior of the vehicle is indicative of a vehicle maneuver that is suspected as being an obstacle avoidance maneuver.

The vehicle maneuver suspected as being an obstacle avoidance maneuver may be detected based on behavioral information that may be obtained from one or more sensors such as an accelerometer, a wheel speed sensor, a vehicle speed sensor, a brake sensor, a steering wheel sensor, an engine sensor, a shock absorber sensor, a driver sensor for sensing one or more physiological parameter of the driver (such as heart beat), or any other sensor—especially any other non-visual sensor.

Step1930 may be followed by step1940 of processing the images of the environment of the vehicle obtained during a vehicle maneuver that is suspected as being the obstacle avoidance maneuver to provide visual information.

The visual information may include one or more robust signatures of one or more images acquired by one or more visual sensors of the vehicle. A example of a robust signature is illustrated in U.S. Pat. No. 8,326,775 which is incorporated herein by reference.

Step1940 may be followed by step1950 of transmitting the visual information to a system that is located outside the vehicle.

Method1900 may also include step1960 of receiving at least one visual obstacle identifier from a remote computer.

Method1900 may include step1970 receiving verification information indicative of whether the maneuver that was suspected being an obstacle avoidance maneuver is actually an obstacle avoidance maneuver. This may be used in the vehicle during step1930.

Methods

1900 and1600 illustrate a learning process. The products of the learning process may be used to detect obstacles—during an obstacle detection process that uses the outputs of the learning process.

The obstacle detection process may include detecting newly detected obstacles that were not detected in the past.

The learning process may continue during the obstacle detection process or may terminate before the obstacle detection process.

FIG. 12 illustratesmethod2100 for detecting an obstacle.

Method

2100 may start bystep2110 of receiving, by an I/O module of a vehicle, a visual obstacle identifier for visually identifying an obstacle; wherein the visual obstacle identifier is generated based on visual information acquired by at least one visual sensor during an execution of at least vehicle maneuver that is suspected as being an obstacle avoidance maneuver. Thus step2110 may include receiving outputs of the learning processes ofmethod1600.

Method

2100 may also includestep2120 of acquiring, by a visual sensor of the vehicle, images of an environment of the vehicle.

Step

2120 may be followed bystep2130 of searching, by a processing circuitry of the vehicle, in the images of the environment of the vehicle for an obstacle that is identified by the visual obstacle identifier.

If finding such an obstacle then step2130 may be followed bystep2140 of responding, by the vehicle, to a detection of an obstacle.

The responding may be based on a driving policy, may be responsive to the mode of controlling the vehicle (any level of automation—starting from fully manual control, following by partial human intervention and ending with fully autonomous).

For example—the responding may include generating an alert perceivable by a human driver of the vehicle, sending an alert to a computerized system located outside the vehicle, handing over the control of the vehicle to a human driver, handing over a control of the vehicle to an autonomous driving manager, sending an alert to an autonomous driving manager, performing an obstacle avoidance maneuver, determining whether the obstacle is a newly detected obstacle, informing another vehicle about the obstacle, and the like.

Determining a Location of a Vehicle

A vehicle, especially but not limited to an autonomous vehicle, has to know its exact location in order to propagate in an efficient manner. For example—the vehicle may receive an obstacle warning that includes the location of the obstacle.

Responding to the obstacle warning may require knowing the exact location of the vehicle. Furthermore—an autonomous driving operation may require a knowledge of the exact location of the vehicle or may be executed in a more optimal manner if the exact location of the vehicle is known. In general, the processing burden associated with autonomous driving decision may be lowered when an autonomous driving manager is aware of the location of the vehicle.

Global positioning systems (GPS) are inaccurate and cannot be solely used for locating the exact location of a vehicle.

There is a growing need to determine the exact location of the vehicle in an efficient manner.

There is a provided an efficient and accurate method for determining the exact location of a vehicle.

The method is image-based in the sense that it is based on images acquired by the vehicle and on a comparison between the acquired images and reference information that was acquired at predefined locations.

The determining of the actual location of the vehicle is of a resolution that is smaller than a distance between adjacent reference images. For example—the resolution may be a fraction (for example between 1-15%, between 20-45%, of between 5-50%) of the distance between adjacent reference images. In the following example it is assumed that the distance between adjacent reference images is about 1 meter and the resolution is about 10 to 40 centimeters.

FIG. 13 illustratesmethod1300 for determining a location of a vehicle. receiving reference visual information that represents multiple reference images acquired at predefined locations.

Method

1300 may start bystep1310 of receiving reference visual information that represents multiple reference images acquired at predefined locations.

The multiple reference images may be taken within an area in which the vehicle is located. The area may be of any size and/or dimensions. For example, the area may include a part of a neighborhood, a neighborhood, a part of a city, a city, a part of a state, a state, a part of a country, a country, a part of a continent, a continent, and even the entire world.

A computerized system and the vehicle may exchange reference visual information in real-time (or near real-time) and/or in a burst mode, where the reference visual information may be provided either periodically and/or when conditions are suitable for a data exchange between the computerized system and the vehicle. For example, a vehicle may be configured to perform regular uploads/downloads of reference visual information when the engine is turned on (or off, with a keep alive battery function to facilitate communication with the computerized system). Alternatively, or in addition, the vehicle may be configured to perform such uploads/downloads when stationary (e.g., parked, or waiting at a traffic light) and a sufficiently reliable wireless connection is detected. Alternatively, or in addition, the uploads/downloads may be triggered by location, where the reference visual information may be downloaded from the computerized system when the vehicle enters an area for which it does not have up-to-date reference visual information.

The reference visual information may include the reference images themselves or any other representation of the reference images.

The reference information may include robust signatures of reference images. The robust signature may be robust to at least one out of noise, lighting conditions, rotation, orientation, and the like—which may provide a robust location detection signatures. Furthermore—using robust signatures may simplify the acquisition of the reference images—as there is no need to acquired reference images in different lighting conditions or other conditions to which the robust signature is indifferent.

The comparison between the reference information and information from the acquired image may be even more efficient when the reference visual information and information regarding the acquired image is represented in a cortex representation.

A cortex representation of an image is a compressed representation of the visual information that is generated by applying a cortex function that include multiple compression iteration. A signature of an image may be a map of firing neurons that fire when a neural network is fed with an image. The map undergoes multiple compression iterations during which popular strings are replaced by shorter representations.

Method

1300 may also includestep1320 of acquiring, by a visual sensor of the vehicle, an acquired image of an environment of the vehicle. It should be noted that multiple images may be acquired by the vehicle—but for simplicity of explanation the following text will refer to a single image. The vehicle may repetitively determine its location—and multiple repetitions of

steps

1320,1330 and1340 may be repeated multiple times—for different images.

Step

1330 may include generating, based on the acquired image, acquired visual information related to the acquired image.

The location determination may be based on visual image regarding static obstacles. Accordingly—each one of the reference visual information and the acquired visual information may include static visual information related to static objects.

The extraction of the static visual information may include reducing from the visual information any visual information that is related to dynamic objects. The extraction of the static information may or may not require object recognition.

The acquired visual information may include acquired static visual information that may be related to at least one static object located within the environment of the vehicle.

Step

1330 may be followed bystep1340 of searching for a selected reference image out of the multiple reference images. The selected reference image may include selected reference static visual information that best matches the acquired static visual information.

Step

1340 may be followed by step1350 of determining an actual location of the vehicle based on a predefined location of the selected reference image and to a relationship between the acquired static visual information and to the selected reference static visual information.

The determining of the actual location of the vehicle may be of a resolution that may be smaller than a distance between the selected reference image and a reference image that may be immediately followed by the selected reference image.

The determining of the actual location of the vehicle may include calculating a distance between the predefined location of the selected reference image and the actual location of the vehicle based on a relationship between at least one value of a size parameter of the at least one static object within the selected reference image and at least one value of the size parameter of the at least one static object within the acquired image.

The determining of the actual location of the vehicle may include determining whether the actual location of the vehicle precedes the predefined location of the selected reference image or precedes the predefined location of the selected reference image.

The method may include subtracting the first distance from the predefined location of the selected reference image when determining that the actual location of the vehicle precedes the predefined location of the selected reference image.

The method may include adding the first distance from the predefined location of the selected reference image when determining that the actual location of the vehicle follows the predefined location of the selected reference image.

Each predefined location may be associated with a single reference image.

Alternatively, each reference image may be associated with multiple images that differ from each other by scale. The images that differ from each other by scale may represent different relative distances between the vehicle and the predefined image—larger scale may be related to smaller distances.

Using images of different scale (or visual information related to images of different scale) may reduce the computational load associated with location determination—as it may eliminate the need to manipulate a single reference image to find an exact math that may determine the distance between the vehicle and the reference location.

Using visual information related to images of different scale may also assist when the visual information is a lossy representation of the images in which a representation of an object of certain scale cannot be easily generated based on a representation of the same object of another scale. The visual information may be a signature of an image may be a map of firing neurons that fire when a neural network is fed with an image. See, for example, the signature of U.S. Pat. Nos. 8,326,775 and 8,312,031, assigned to common assignee, which are hereby incorporated by reference.

FIG. 14 illustrates an example of avehicle VH11402 that acquires an image I(k)1404 and tries to determine its location based on analysis of visual information it generates from image I(k) and multiple reference images RI(1,1)-RI(j,1)1402(1)-1402(j) and RI(j+1,1)-RI(j)1402(j+1,1)-1402(J,1). The front of the vehicle is positioned between the predefined locations associated with reference images RI(j) and RI(j+1)—and one of these images should be associated with reference visual information that best matches the visual information associated with image I(k). The distance between consecutive predefined locations isDRI1422—and the location resolution is finer than DRI.

FIG. 15 illustrates an example of avehicle VH11402 that acquires an image I(k)1404 and tries to determine its location based on analysis of visual information (1406) it generates from image I(k) and visual information related to reference images. InFIG. 15 each predefined location is associated with a plurality of reference images—with a plurality (p) of reference visual information per reference location —1408(1,1)-1408(j,1),1408(j+1,1)-1408(J,1),1409(1,p)-1409(j,p), and1409(j+1,p)-1409(J,p).

The front of the vehicle is positioned between the j′th and (j+1)′th predefined locations and the vehicle may determine the best matching predefined location and also the distance from the predefined locations based on the p different reference visual information associated with one of these predefined locations.

FIGS. 17 and 18 illustrate an example of an image and static visual information.

FIG. 17 illustrates an example of an image acquired by a vehicle. The image shows two

lanes

1411 and1412 of a road,right building1413,

vehicles

1414 and1415, first tree1416, table1417,second tree1418, cloud1419,rain1420, leftbuilding1421,shrubbery1422,child1423 andball1424.

The dynamic objects of this image include

vehicles

1414 and1415, cloud1419,rain1420,shrubbery1422,child1423 andball1424. These dynamic objects may be represented by visual information (such as signatures) that describe them as dynamic—and visual information related to these dynamic objects may be removed from the visual information of the images to provide static visual information.

FIG. 18 illustrates the removal (blocks1432) of dynamic visual information from thevisual information1430 to provide staticvisual information1431.

Visual information

1430 may be a map of firing neurons of a network that fired when the neural network was fed with image I(k)- or may be a compressed representation (for example a cortex representation) of such map.

Triggering Human Intervention

FIG. 19 illustrates method1450.

Method1450 may be aimed to find at least one trigger for human intervention in a control of a vehicle.

Method1450 may start bystep1452 of receiving, from a plurality of vehicles, and by an I/O module of a computerized system, visual information acquired during situations that are suspected as situations that require human intervention in the control of at least one of the plurality of vehicles.

Step

1452 may be followed bystep1456 of determining, based at least on the visual information, the at least one trigger for human intervention. The at least one trigger is represented by trigger visual information.

Step

1456 may include at least one out of

- Determining a complexity of the situation.
- Determining a danger level associated with the situation.
- Determining in response to statistics of maneuvers executed by different vehicles during a same situation that is suspected as a situation that requires human intervention in the control of at least one of the plurality of vehicles.
- Determining based on generated or received movement information of entities included in the visual information. The movement information may represent entity movement functions of the entities.
- The determining may include estimating, based on the entity movement functions, a future movement of the entities.

The determining ofstep1456 may be executed in an unsupervised manner or a supervised manner.

The determining ofstep1456 is responsive to at least one human intervention policy of the at least one vehicle. The human intervention policy of a vehicle may differ from the human intervention policy of another vehicle.

A human intervention policy of a vehicle may define certain criteria human intervention such as the danger level of a situation, the complexity of the situation (especially complexity of maneuvers required to overcome the situation), the potential damage that may result to the vehicle, driver or surroundings due to the situation. A human intervention policy of a vehicle may also define certain situations that require human intervention—such as specific locations (for example near a cross road of a school), or specific content (near a school bus), and the like.

A trigger may be determined duringstep1456 when the situation or one of the attributes of the situation (danger level, location, combination of location and time) match those of the human intervention policy.

Situations that are suspected as requiring human intervention in the control of a vehicle may be viewed as situations that are complex and/or dangerous and/or require changes in the behavior of a vehicle. These situations may result from an obstacle or may result from other reasons.

Situations that are suspected as requiring human intervention in the control of a vehicle may be automatically detected by processing images obtained while a vehicle performs maneuvers that are suspected as resulting from situations that are suspected as requiring human intervention in the control of a vehicle.

Situations that are suspected as requiring human intervention in the control of a vehicle may be detected based on behavioral information regarding behavior of the plurality of vehicles, during an execution of the maneuvers that are suspected as resulting from situations that are suspected as requiring human intervention in the control of a vehicle.

The behavioral information may represent the behavior (speed, direction of propagation, and the like) of the entire vehicle and/or may represent the behavior of parts or components of the vehicle (for example damping a shock by a shock absorber, slowing the vehicle by the breaks, turns of the steering wheel, and the like), and/or may represent the behavior of the driver (the manner in which the driver drives), and the like.

After a situation is tagged as requiring human intervention in the control of a vehicle—the vehicle may trigger human intervention when this situation is visually detected. For example—some events (or some types of events) may be defined (even based on previous recognition of a situation as requiring human intervention in the control of a vehicle) as requiring human intervention. For example—an image of a cross road located near a school, an image of a school bus that transports children, an image of a bar and drunken pedestrians near the bar, an image of multiple people that are near the road, and the like may be flagged as situation as requiring human intervention in the control of a vehicle

A determination of what constitutes a vehicle maneuver suspected as requiring human intervention in the control of a vehicle may be detected in a supervised manner or in an unsupervised manner.

The vehicle maneuver suspected as related to a situation that requires human intervention in the control of a vehicle may involve a change (especially a rapid change) in the direction and/or speed and/or acceleration of the vehicle, a deviation from a lane, a deviation from a previous driving pattern followed at attempt to correct the deviation, and the like.

For example—when applying supervised learning of obstacles—the computerized system may be fed with examples of vehicle behaviors that require human intervention in the control of a vehicle.

For example—a fast and significant change in the speed and direction of a vehicle at a linear road path may indicate that the vehicle encountered a situation requiring human intervention in the control of a vehicle

Step

1456 may be followed bystep1460 of transmitting to one or more of the plurality of vehicles, the trigger visual information.

FIG. 20 illustrates amethod1490 for detecting situation that requires human intervention in the control of a vehicle. The method may be executed by a vehicle.

Method

1490 may start by

steps

1491 and1492.

Step

1491 may include sensing, by a non-visual sensor of a vehicle, a behavior of a vehicle. The non-visual sensor may be an accelerometer, a shock absorber sensor, a brakes sensor, and the like.

Step

1492 may include acquiring, by a visual sensor of the vehicle, images of an environment of the vehicle.

Steps

1491 and1492 may be followed bystep1493 of determining, by a processing circuitry of the vehicle, whether the behavior of the vehicle is indicative of a vehicle maneuver that is related to a situation that is suspected as requiring human intervention in the control of a vehicle.

The vehicle maneuver related to a situation that is suspected as requiring human intervention in the control of a vehicle may be detected based on behavioral information that may be obtained from one or more sensors such as an accelerometer, a wheel speed sensor, a vehicle speed sensor, a brake sensor, a steering wheel sensor, an engine sensor, a shock absorber sensor, a driver sensor for sensing one or more physiological parameter of the driver (such as heart beat), or any other sensor—especially any other non-visual sensor.

The vehicle maneuver related to a situation that is suspected as requiring human intervention in the control of a vehicle may involve a change (especially a rapid change) in the direction and/or speed and/or acceleration of the vehicle, a deviation from a lane, a deviation from a previous driving pattern followed at attempt to correct the deviation, and the like.

Step

1493 may be followed bystep1494 of processing the images of the environment of the vehicle obtained during a vehicle maneuver related to a situation that is suspected as requiring human intervention in the control of a vehicle to provide visual information.

Step

1494 may be followed bystep1495 of transmitting the visual information to a system that is located outside the vehicle.

Method

1490 may also includestep1496 of receiving at least one visual obstacle identifier from a remote computer.

Method

1490 may includestep1497 receiving verification information indicative of whether the maneuver related to a situation that is suspected as requiring human intervention in the control of a vehicle maneuver is actually related to a situation that actually requires human intervention in the control of a vehicle. This may be used in the vehicle duringstep1493.

Methods

1450 and1490 illustrate a learning process. The products of the learning process may be used to detect situations that require human intervention in the control of a vehicle—during a vehicle control process that uses the outputs of the learning process.

The learning process may continue during the vehicle control process or may terminate before the vehicle control process.

FIG. 21 illustratesmethod1470.

Method

1470 may start bystep1472 of receiving, by an I/O module of a vehicle, trigger visual information for visually identifying situations that require human intervention in the control of a vehicle. The visual obstacle identifier is generated based on visual information acquired by at least one visual sensor during an execution of at least vehicle maneuver that is suspected as being an obstacle avoidance maneuver. Thus step1472 may include receiving outputs of the learning processes of method1450.

Method

1470 may also includestep1474 of acquiring, by a visual sensor of the vehicle, images of an environment of the vehicle.

Step

1474 may be followed bystep1476 of searching, by a processing circuitry of the vehicle, images of the environment of the vehicle for a situation that is identified by the trigger visual information.

If finding such a situation then step1476 may be followed bystep1478 of triggering human intervention in the control of a vehicle. This may include alerting the driver that human intervention is required, handing the control of the vehicle to the driver- and the like.

Step

1478 may include, for example—generating an alert perceivable by a human driver of the vehicle, sending an alert to a computerized system located outside the vehicle, handing over the control of the vehicle to a human driver, handing over a control of the vehicle from an autonomous driving manager, sending an alert to an autonomous driving manager, informing another vehicle about the trigger, and the like.

FIGS. 22-27 illustrate example of situations that are suspected as requiring human intervention in the control of a vehicle.

FIG. 22 illustratesfirst vehicle VH11801 as stopping (position1501) in front of apuddle1506 and then passing the puddle (may drive straight or change its direction till ending the maneuver atpoint1502. The maneuver may be indicative that passing the puddle may require human intervention.

Visual information acquired betweenpositions1501 and1502 are processed duringstep1494.

FIG. 23 illustratesfirst vehicle VH11801 as sensing

pedestrians

1511 and1512. These pedestrians either are associated with temporal signatures that illustrate their movement (1511′ and1512′) and/or the vehicle may sense the movements of the pedestrians—and the vehicle may send (based on the movement) the future movement and the future location (1511″ and1512″) of the pedestrians and perform a maneuver that may include altering speed (for example stopping at point1513) and/or bypassing the pedestrians (maneuver1515).

Visual information acquired betweenpositions1513 and1514 (end of the maneuver) are processed duringstep1494.

FIG. 24 illustratesfirst vehicle VH11801 as sensing parked vehicles PV11518 and PV21519 that part on both sides of a double-lane bi-directional road, that require the first vehicle to perform acomplex maneuver1520 that includes changing lanes and changing direction relatively rapidly.

Visual information acquired between positions1516 (beginning of the maneuver) and1517 (end of the maneuver) are processed duringstep1494.

FIG. 25 illustratesfirst vehicle VH11801 as stopping (position1522) in front of wet segment of the road on which rain1521 (from cloud1522) falls. The stop (at location1522) and any further movement after moving to another part of the road may be regarded as amaneuver1523 that is indicative that passing the wet segment may require human intervention.

Visual information acquired between position1522 (beginning of the maneuver) and the end of the maneuver are processed duringstep1494.

FIG. 26 illustratesfirst vehicle VH11801 as stopping (position1534) in front of a situation that may be labeled as a packing or unpacking situation—atrack1531 is parked on the road, there is anopen door1532, and a pedestrian1533 carries luggage on the road. Thefirst vehicle1801 bypasses the truck and the pedestrian between

locations

1534 and1535 duringmaneuver1539. The maneuver may be indicative that a packing or unpacking situation may require human intervention.

Visual information acquired between

positions

1534 and1535 are processed duringstep1494.

FIG. 27 illustratesfirst vehicle VH11801 as turning away (maneuver1540) from the road when sensing that it faces asecond vehicle VH21802 that moves towardsVH11801. Maneuver1540 may be indicative that such as potential collision situation may require human intervention.

Visual information acquired betweenpositions1541 and1542 (start and stop of maneuver1540) are processed duringstep1494.

Tracking after an Entity

There may be beneficial to detect and object in a robust and an efficient manner by considering the temporal behavior of the entity and also its spatial signature. Using signatures of multiple type may increase the reliability of the detection, one signature may verify the other signature. In addition—one signature may assist in detecting the entity when the other signature is missing or of a low quality. Furthermore—estimating the future movement of an entity-based on the spatial signature of the entity (or of those of entities of the same type) may enable to predict the future effect that this entity may have on the vehicle and allow to determine the future driving pattern of the vehicle in advance.

The temporal signature may be compresses—thus saving memory and allowing a vehicle to allocate limited memory resources for tracking each entity—even when tracking multiple entities.

FIG. 28 illustrates amethod2600 for tracking after an entity.

Method

2600 may include

steps

2602,2604,2606 and2608.

Tracking, by a monitor of a vehicle, a movement of an entity that appears in various images acquired during a tracking period. S2602.

Generating, by a processing circuitry of the vehicle, an entity movement function that represents the movement of the entity during the tracking period. S2604.

Generating, by the processing circuitry of the vehicle, a compressed representation of the entity movement function. S2606.

Responding to the compressed representation of the entity movement function. S2608.

The compressed representation of the entity movement function may be indicative of multiple properties of extremum points of the entity movement function.

The multiple properties of an extremum point of the extremum points, may include a location of the extremum point, and at least one derivative of the extremum point.

The multiple properties of an extremum point of the extremum points may include a location of the extremum point, and at least two derivative of at least two different orders of the extremum point.

The multiple properties of an extremum point of the extremum points, may include a curvature of the function at a vicinity of the extremum point.

The multiple properties of an extremum point of the extremum points, may include a location and a curvature of the function at a vicinity of the extremum point.

Method

2600 may include acquiring the images by a visual sensor of the vehicle. S2601.

At least one image of the various images may be acquired by an image sensor of another vehicle.

Step

2608 may include at least one out of:

- Storing, in a memory unit of the vehicle, the compressed representation of the entity movement function.
- Transmitting the compressed representation of the entity movement function to a system that may be located outside the vehicle.
- Estimating, by a processing circuitry of the vehicle, a future movement of the entity, based on the compressed representation of the entity movement function.
- Generating a profile of the entity, by a processing circuitry of the vehicle, based on the compressed representation of the entity movement function.
- Predicting an effect of a future movement of the entity on a future movement of the vehicle, wherein the predicting may be executed by a processing circuitry of the vehicle, and may be based on the compressed representation of the entity movement function.
- searching for a certain movement pattern within the movement of the entity, by a processing circuitry of the vehicle, based on the compressed representation of the entity movement function.

Method

2600 may include receiving a compressed representation of another entity movement function, the other entity movement function may be generated by another vehicle and may be indicative of the movement of the entity during at least a subperiod of the tracking period.52610.

Method

2600 may include amending the compressed representation of the entity movement function based on the compressed representation of the other entity movement function.52612.

Method

2600 may include determining a duration of the tracking period.52603. The duration may be determine based on the certainty level of the prediction, based on memory and/or computational resources allocated to the tracking, and the like.

FIG. 29 illustrates examples of entity movement functions. One axis is time. Other axes are spatial axes (x-axis, y-axis and the like).

FIG. 30 illustratesmethod2630.

Method

2630 may include

steps

2631,2632,2634 and2636.

Method

2630 may include calculating or receiving an entity movement function that represents a movement of the entity during a tracking period.52632.

Searching, by a search engine, for a matching reference entity movement function.52634

Identifying the entity using reference identification information that identifies a reference entity that exhibits the matching reference entity movement function. S2636.

The reference identification information may be a signature of the entity.

The method may include acquiring a sequence of images by an image sensor; and calculating the entity movement based on the sequence of images. S2631

FIG. 31 illustrates amethod2640.

Method

2630 may include

steps

2642,2644,2646 and2648.

Method

2640 may start by calculating or receiving multiple entity movement functions that represent movements of multiple entities. S2642.

Clustering the multiple entity movement functions to clusters. S2644.

For each cluster, searching, by a search engine, for a matching type of reference entity movement functions. S2646.

Identifying, for each cluster, a type of entity, using reference identification information that identifies a type of reference entities that exhibits the matching type of reference entity movement functions. S2648.

FIG. 32 illustrates amethod2650.

Method

2650 may include

steps

2652,2654 and2656.

Method

2650 may start by calculating or receiving (a) an entity movement function that represents a movement of an entity, and (b) a visual signature of the entity. S2652.

Comparing the entity movement function and the visual signature to reference entity movement functions and reference visual signatures of multiple reference objects to provide comparison results. S2654.

Classifying the object as one of the reference objects, based on the comparison results. S2656.

FIG. 33 illustrates amethod2660.

Method

2660 may include

steps

2662,2664,2666 and2668.

Step

2662 may include calculating or receiving an entity movement function that represents a movement of an entity.

Step

2664 may include comparing the entity movement function to reference entity movement functions to provide comparison results.

Step

2666 may include classifying the object as a selected reference object of the reference objects, based on the comparison results.

Step

2668 may include verifying the classifying of the object as the selected reference object by comparing a visual signature of the object to a reference visual signature of the reference object.

FIG. 34 illustrates amethod2670.

Method

2670 may include

steps

2672,2674,2676 and2678.

Step

2672 may include calculating or receiving a visual signature of the object.

Step

2674 may include comparing a visual signature of the object to reference visual signatures of multiple reference objects to provide comparison results.

Step

2644 may include classifying the object as a selected reference object of the reference objects, based on the comparison results.

Step2678 may include verifying the classifying of the object as the selected reference object by comparing an entity movement function that represents a movement of the entity to a reference entity movement functions to provide comparison results.

FIG. 35 illustrates amethod2680.

Method

2680 may include

steps

2682,2684 and2686.

Method

2680 is for generating a signature of an object.

Step

2682 may include calculating or receiving a visual signature of the object.

Step

2684 may include calculating or receiving an entity movement function that represents a movement of the object.

Step

2686 may include generating a spatial-temporal signature of the object that represents the visual signature and the entity movement function of the object.

FIG. 36 illustrates a method2700.

Method2700 may include steps2702,2704 and2706.

Method2700 is for driving a first vehicle based on information received from a second vehicle.

The exchange of information—especially compact and robust signatures between vehicles may improve the driving of the vehicle and at a modest cost—especially when the vehicles exchange compact signatures between themselves.

Step2702 may include receiving, by the first vehicle, acquired image information regarding (a) a signature of an acquired image that was acquired by the second vehicle, (b) a location of acquisition of the acquired image.

Step2704 may include extracting, from the acquired image information, information about objects within the acquired image.

Step2706 may include preforming a driving related operation of the first vehicle based on the information about objects within the acquired image.

The acquired image information regarding the robust signature of the acquired image may be the robust signature of the acquired image.

The acquired image information regarding the robust signature of the acquired image may be a cortex representation of the signature.

The method may include acquiring first vehicle images by the first vehicle; extracting, from the first vehicle acquired images, information about objects within the first vehicle acquired images; and preforming the driving related operation of the first vehicle based on the information about objects within the acquired image and based on the information about objects within the first vehicle acquired images.

The image information may represent data regarding neurons of a neural network, of the second vehicle, that fired when the neural network was fed with the acquired image.

The extracting, of the information about objects within the acquired image may include (a) comparing the signature of the acquired image to concept signatures to provide comparison results; each concept signature represents a type of objects; and (b) determining types of objects that may be included in the acquired image based on the comparison results.

FIG. 37 illustrates one vehicle that updates another vehicle—especially with objects that are not currently seen by the other vehicle—thereby allowing the other vehicle to calculate his driving path based (also) on the object it currently does not see.

Concept Update

A concept structure includes multiple signatures that are related to each other and metadata related to these signatures.

The concepts may be updated by removing signatures—and even parts of the signatures that are more “costly” to keep than maintain.

The “cost” may represent one or more factors such as false detection probability, a robustness of the concept, an accuracy of the concept, and the like.

One or more vehicles may decide to update a concept—and then send the update (or an indication about the update) to other vehicles—therefore improving the detection process by using updated concepts.

FIG. 38 illustrates amethod2710.

Method

2710 may include

steps

2712,2714 and2726.

Method

2710 is for concept update.

Method

2710 may include:

- Step2712 of detecting that a certain signature of an object causes a false detection (see, for exampleFIGS. 39 and 40). The certain signature belongs to a certain concept structure that may include multiple signatures. The false detection may include determining that the object may be represented by the certain concept structure while the object may be of a certain type that may be not related to the certain concept structure. For example—a concept of a pedestrian may classify (by error) a mail box as a pedestrian.
- Step2714 of searching for an error inducing part (see, for exampleFIGS. 41-43) of the certain signature that induced the false detection.
- Step2715 of determining whether to remove the error inducing part of the certain signature. This may involve calculating a cost related to a removing the error inducing part from the concept structure—and removing the error inducing part when the cost may be within a predefined range.
- Step2726 of removing (see, for exampleFIG. 44) from the concept structure the error inducing part to provide an updated concept structure.

Each signature may represent a map of firing neurons of a neural network.

Step

2714 may include:

- Generating or receiving a test concept structure that includes (a) first signatures of images that may include one or more objects of the certain type (that should not belong to the concept), and (b) second signatures of second images that may include one or more objects of a given type that may be properly associated with the concept structure. The images of both types may be selected in any manner—for example may be randomly selected.
- Comparing the certain signature to the test concept structure to provide matching first signatures and matching second signatures.
- Comparing the matching first signatures and matching second signatures to find parts that causes false errors and parts that result in positive detection.
- Defining the error inducing part of the certain signature based on an overlap between the matching parts of the certain signature.

The updated concept may be shared between vehicles.

Situation Aware Processing

Reference information such as but not limited to signatures, cluster structures and the like may be related to a vast number of situations. It has been found that situation aware processing may be much more effective than processing that is not situation aware.

Situation aware processing includes various steps, some of which may include detecting a situation and then performing a situation related processing based on the situation. The situation related processing may include, for example, object detection, object behavior estimation, responding to an outcome of the object detection, responding to an outcome of an object behavior estimation, and the like.

Various examples of responses to object detection and/or responding to an outcome of an object behavior estimation are provided inFIGS. 3A-44. For example—the responding may include at least one out of (i) performing an obstacle avoidance movement in response to the outcome of the object detection and/or the object behavior estimation, (ii) autonomously driving an autonomous vehicle in response to the outcome of the object detection and/or the object behavior estimation, (iii) assisting a driver in response to the outcome of the object detection and/or the object behavior estimation, (iv) determining which entity (human or non-human) will control the autonomous vehicle in response to the outcome of the object detection and/or the object behavior estimation, (v) communicate with other vehicles in response to the outcome of the object detection and/or the object behavior estimation, and the like.

Various examples of a situation may be at least one of (a) a location of the vehicle, (b) one or more weather conditions, (c) one or more contextual parameters, (d) a road condition, (e) a traffic parameter.

Various examples of a road condition may include the roughness of the road, the maintenance level of the road, presence of potholes or other related road obstacles, whether the road is slippery, covered with snow or other particles.

Various examples of a traffic parameter and the one or more contextual parameters may include time (hour, day, period or year, certain hours at certain days, and the like), a traffic load, a distribution of vehicles on the road, the behavior of one or more vehicles (aggressive, calm, predictable, unpredictable, and the like), the presence of pedestrians near the road, the presence of pedestrians near the vehicle, the presence of pedestrians away from the vehicle, the behavior of the pedestrians (aggressive, calm, predictable, unpredictable, and the like), risk associated with driving within a vicinity of the vehicle, complexity associated with driving within of the vehicle, the presence (near the vehicle) of at least one out of a kindergarten, a school, a gathering of people, and the like. A contextual parameter may be related to the context of the sensed information-context may be depending on or relating to the circumstances that form the setting for an event, statement, or idea.

FIG. 45 illustrates an example ofmethod3000.

Method

3000 may start bystep3009 of sensing or receiving first sensed information. The first sensed information may be any type of information (visual, radar, sonar, non-visual, and the like).

The first sensed information may be sensed by at least one first sensor of a vehicle.

The first sensed information may be sensed by a sensor that does not belong to a vehicle.

Step

3009 may be followed bystep3014 of detecting a situation based on first sensed information.

Step

3014 may be followed bystep3018 of selecting, from reference information, a situation related subset of the reference information. The situation related subset of the reference information is related to the situation.

Step

3014 of detection of the situation may include object detection andstep3018 of the selection may be responsive to the detected objects. Alternatively,step3014 may not include object detection and/orstep3018 may not be based on a detection or lack of detection of a certain object. For example—if a vehicle is at a certain inclination than step2018 may include selecting only a part of sensed information based on the inclination (for example if the front vehicle is in a downwards inclination than only a certain lower part of an image sensor of the vehicle may be processed—assuming that an upper part may include irrelevant information—such a clouds or the sky.

For example—step3018 may include limiting, without recognizing objects, a processing at certain areas of the picture. (For example—when the vehicle is estimated to be ahead of a T-junction, an only relevant area to process is located at the lower part (for example lower half—or any other part—depending at least in part on the inclination and the field of view of the sensors acquiring the image). The selection may not require identify any objects specifically. The limitation of the processing area is based on situation identification.

The selection may be responsive to statistical information, probabilistic information or any other information related to the chances of an appearance of a situation significant information in different parts of the sensed information—for example the appearance of a pedestrian at different locations near a roundabout, within a roundabout, and the like.

The statistical information, probabilistic information or any other information related to the chances of an appearance of a situation significant information in different parts of the sensed information may be used to adjust, fine tune or otherwise adapt the situation related processing. For example—changing detection rules, changing similarity tests, altering signatures match criteria, changing threshold of similarity for inclusion of a signature within a concept, and the like

The situation related subset of the object related reference information is a subset of reference information required for performing a situation related processing operation that may be relevant to the situation. The reference information may be, for example, information that may be compared to sensed information in order to detect objects, detect movement patterns, and the like. The comparing is merely one example of how the reference information can be used.

Various examples of a situation related subset of the reference information may be related to at least one out of (a) a location of the vehicle, (b) one or more weather conditions, (c) one or more contextual parameters, (d) a road condition, (e) a traffic parameter, (f) a potential risk, and the like.

For example—the situation related subset may include at least one out of (a) signatures and metadata of a cluster structure of roundabouts, (b) signatures and metadata of a cluster structure of a cross road located at a certain city, (c) signatures and metadata of a cluster structure of a highway acquired at sunset, (d) signatures and metadata of a cluster structure of a specific road segment, (e) signatures and metadata of a cluster structure of a specific road segment taken at a certain weather condition, (f) signatures and metadata of a cluster structure of certain roundabout acquired at rush hour, and the like.

The reference information may be stored in a memory unit of the vehicle and the selecting may include informing a processor the locations of the situation related subset, allocating a higher cache or other data retrieval priority to the subset, and the like.

At least a part of the subset may be stored outside the vehicle and the selecting may include retrieving the at least part of the subset to the vehicle.

The retrieval of the situation related subset may include retrieving it to a cache or other memory unit (for example a high priority memory unit) allocated to the processor or included in the processor.

The retrieval may include ignoring reference information not related to the situation—reference information that differs from the situation related subset.

The retrieval of only the situation related subset reduces memory resource consumption—as the situation related subset and not the entire reference information is loaded to the vehicle and/or to the processor and/or to the cache and/to a high priority memory unit.

Method

3000 may also includestep3020 of sensing second sensed information.

The second sensed information is sensed at a second period.

The second period may follow the detecting of the situation, may precede the detection of the situation, may partially overlap (or fully overlap) the detecting of the situation, may follow the detecting of the situation, may precede the detecting of the situation, may partially overlap (or fully overlap) the detecting of the situation.

The first sensed information may be included in the second sensed information, may partially overlap the second sensed information, may not overlap the second sensed information, may differ by type from the first sensed information, and the like.

For example an image that is processed to detect a situation may also be used for image processing, may be a part of images that are processed to detect movement of objects, and the like.

The second sensed information may be sensed by at least one sensor of a vehicle.

The second sensed information may be sensed by a sensor that does not belong to a vehicle.

The first and second sensed information may be sensed by the same sensor or sensors, may be sensed by completely different sensors, may be sensed in part by a same sensor.

Steps

3020 and3018 may be followed bystep3024 of performing, by a situation related processing unit, a situation related processing. The situation related processing is based on the situation related subset of the reference information and on the second sensed information sensed by the at least one vehicle sensor at a second period.

The situation related processing may include object detection and/or object behavior estimation.

Step

3024 may be executed by any of the method illustrated in the specification.

For example—step3024 may include calculating one or more second signatures of the second sensed information; and searching, out of the concept structures related to the situation, for at least one matching concept structure that matches the one or more second signatures, and determining at least an identity of a detected object based on metadata of the at least one matching concept structure.

Step

3024 may include, for example, object detection, object behavior estimation, and the like.Step3024 may include, for example applying any type of machine learning algorithms, not necessarily signatures. e.g. Deep Learning where models are loaded, and input is a direct “pixel information” of the sensor.Step3024 may be executed without signature calculation.

Step

3014 of detecting the situation and step3024 may be executed at the same manner—using the same algorithms, applying a same type of operations, exhibit a same or similar levels of complexity, exhibit a same or similar consumption of memory resources, exhibit a same or similar consumption of processing resources, and the like.

Alternatively—step3014 of detecting the situation and step3024 may differ from each other by at least one out of manner of execution—for example may differ from each other by at least one out of—algorithms, type of operations, levels of complexity, consumption of memory resources, consumption of processing resources, and the like.

Step

3024 may be executed by searching matching concept structures andstep3014 may be executed without searching for matching concept structures.

Step

3014 may be executed calculating signatures of a type used in the object related processing.

Step

3024 may be followed bystep3030 of responding to the outcome of the object related processing.

Various examples of a responses to object detection and/or responding to an outcome of an object behavior estimation are provided inFIGS. 3A-44. For example—the responding may include at least one out of (i) performing an obstacle avoidance movement in response to the outcome of the object detection and/or the object behavior estimation, (ii) autonomously driving an autonomous vehicle in response to the outcome of the object detection and/or the object behavior estimation, (iii) assisting a driver in response to the outcome of the object detection and/or the object behavior estimation, (iv) determining which entity (human or non-human) will control the autonomous vehicle in response to the outcome of the object detection and/or the object behavior estimation, the determining may be responsive to at least one autonomous driving rule, (v) communicate with other vehicles in response to the outcome of the object detection and/or the object behavior estimation, and the like.

Method

3000 may be executed by different vehicles that cooperate with each other—the cooperation may be made in a distributed manner, in a hierarchical or non-hierarchical manner. For example—one vehicle may process first sensed information from another vehicle, two or more vehicle may share resources to executestep3024, and the like.

FIG. 46 illustrates an example ofmethod3001.

Method

3001 ofFIG. 46 differs frommethod3000 ofFIG. 45 by havingstep3010 instead ofstep3009.Step3010 includes sensing the first sensed information by at least one vehicle sensor during the first period. The at least one vehicle sensor belongs to the same vehicle that includes the one or more second sensors that sensed the second sensed information.

Step

3010 is followed bystep3014.Step3014 is followed bystep3018.

Steps

3020 and3014 are followed bystep3018.Step3018 is followed bystep3024.Step3024 is followed bystep3030.

The at least one first vehicle sensor may differ from the one or more second sensors. The at least one first vehicle sensor may be the one or more second sensors. There may be a partial overlap between the at least one first vehicle sensor and the one or more second sensors.

FIG. 47 includes three timing diagrams3071,3072 and3073 that illustrates various timing relationships between thefirst period3061,situation detection3062 and thesecond period3062.

Timing diagram3071 illustrates thesituation detection3062 as following thefirst period3061 and preceding thesecond period3062—without timing overlap.

Timing diagram3071 illustrates thesituation detection3062 as following thefirst period3061. Thesecond period3062 partially overlaps thefirst period3061 and overlaps theentire situation detection3062.

Timing diagram3073 illustrates thesituation detection3062 as following thefirst period3061. Thesecond period3062 and thefirst period3061 start together and thesecond period3061 precedes thesituation detection3062.

FIG. 48 illustrates an example ofmethod3003.

Method

3003 ofFIG. 48 differs frommethod3000 ofFIG. 45 by having step3041 instead ofstep3009, and havingstep3050 instead ofstep3020.

Step3041 includes sensing first sensed information that is first visual information by a camera during a first period.

Step

3050 includes sensing second sensed information that is second visual information by the camera.

Step3041 is followed bystep3014.Step3014 is followed bystep3018.

Steps

3050 and3014 are followed bystep3018.Step3018 is followed bystep3024.Step3024 is followed bystep3030.

FIG. 49 illustrates an example ofmethod3003.

Method

3003 ofFIG. 49 differs frommethod3000 ofFIG. 45 by having step3042 instead ofstep3009, and havingstep3050 instead ofstep3020.

Step3042 includes sensing non-visual information that is first sensed information by a non-visual sensor that during a first period.

Step

3051 includes sensing second sensed information that is second visual information by the camera.

Step3042 is followed bystep3014.Step3014 is followed bystep3018.

Steps

FIG. 50 illustrates an example ofmethod3004.

Method

3003 ofFIG. 49 differs frommethod3000 ofFIG. 45 by having step3049 of uploading the situation related subset of the reference information to a situation related processing unit.

Step

3010 is followed bystep3014.Step3014 is followed bystep3018.Step3018 is followed bystep3019.

Steps

3019 and3020 are followed bystep3018.Step3018 is followed bystep3024.Step3024 is followed bystep3030.

Computational resources and/or memory resources can further be saved when the second sensed information is segmented to portions and the processing of the portions is based on their relevancy to the situation.

Thus—less resources are allocated to the processing of irrelevant second sensed information segments, and such irrelevant portions may be ignored of.

There may be more than two relevancy classes and the allocation of the resources may be based on the relevancy level.

The relevancy may also be determined based on the situation and at least one additional parameters—such as the type of the object that is being detected and/or the type of object that its behavior is being monitored.

Any of the mentioned below methods may include determining, based on the situation, a relevancy value for each portion out of multiple portions of the second sensed information.

The determining may be followed by performing the situation related processing based on the relevancy value of the each portion.

The determining of the relevancy value may include finding, for a certain type of object, a relevant portion and an irrelevant portion. The performing of the situation related processing is executed without searching for the certain type of object within the irrelevant portion.

The determining of relevancy value may be followed by determining a resolution of signatures or concept structures based on the relevancy value of the each portion.

The preforming of the situation related processing within the irrelevant portion may be executed without using concept structures that describe the object of the certain type.

FIG. 51 illustrates an example ofmethod3005.

Method

3005 may start bystep3009 of sensing or receiving first sensed information. The first sensed information may be any type of information (visual, radar, sonar, non-visual, and the like).

Step

Method

3000 may also includestep3020 of sensing second sensed information by the at least one vehicle sensor at a second period.

Step

3030 andstep3009 may be followed bystep3051 of determining, based on the situation, a relevancy value for each portion out of multiple portions of the second sensed information.

Steps

3051 and3018 may be followed bystep3052 of performing, by a situation related processing unit, a situation related processing to find detected objects, wherein the situation related processing is based on the situation related subset of the reference information and on the second sensed information sensed by the at least one vehicle sensor at a second period. The processing is based on the relevancy value of each portion.

Step

3052 may be executed by any of the method illustrated in the specification—but is based on the relevancy value of each portion.

For example—step3052 may include calculating one or more second signatures of the second sensed information; and searching, out of the concept structures related to the situation, for at least one matching concept structure that matches the one or more second signatures, and determining at least an identity of a detected object based on metadata of the at least one matching concept structure.

The length of the signatures may be based on the priority of each portion. For example—shorter or lower resolution signatures may be allocated to lower priority portions. The size of the concepts may be based on the priority of each portion.

Step

3052 may include, for example, object detection, object behavior estimation, and the like.

Step

3014 of detecting the situation and step3052 may be executed at the same manner—using the same algorithms, applying a same type of operations, exhibit a same or similar levels of complexity, exhibit a same or similar consumption of memory resources, exhibit a same or similar consumption of processing resources, and the like.

Step

3024 may be followed bystep3052 of responding to the outcome of the object related processing.

Method

3005 may be executed by different vehicles that cooperate with each other—the cooperation may be made in a distributed manner, in a hierarchical or non-hierarchical manner. For example—one vehicle may process first sensed information from another vehicle, two or more vehicle may share resources to executestep3024, and the like.

FIG. 52 illustrates an example ofmethod3006.

Method

3006 differs from method51 ofFIG. 51 by determining the priority per a combination of second sensed information portions and object types. One image portion may be relevant to one object type and irrelevant for another object type. One image portion may be relevant to multiple object types and irrelevant to one or more object types.

The relevancy of portion to a type of object may be determined in any manner-supervised or unsupervised.

FIG. 53 illustrates animage3080 of a roundabout.Image3080 includes include acurved road3072 within the roundabout, an ring shapedzone3073 that surrounds theinner circle3074 of the roundabout,sand3078 and few bushes orplants3075 located in theinner circle3074,various trees3076 that surround the roundabout, apavement3075 that surrounds the roundabout, and avehicle3071 that is about to enter the roundabout.

FIG. 54 illustrates relevant and irrelevant regions withinimage3080.

Assuming that vehicles are not expected to pass through theinner circle3074 of the roundabout,sand3078 and few bushes orplants3075 located in theinner circle3074—theinner circle3074 and the elements located within the inner circle may be deemed irrelevant for searching for vehicles—as indicated by region3082.

Assuming that pedestrians and vehicle do not fly—the sky part of the image may be regarded as irrelevant for pedestrians and vehicle—as can be noted inregion3081.

Accordingly—the situation related processing may not search for vehicles in regions3082 and3081- and may not retrieve reference concept structures related to vehicles, may not search for matching concept structures from reference concept data structures associated with vehicles, may use, when searching these regions, vehicle signatures of less resolution.

The same applies for searching pedestrians inregion3081.

FIG. 55 illustrates an example of an image acquired by a vehicle. The image shows two

lanes

1411 and1412 of a road,right building1413,

vehicles

It is assumed that vehicles should not be found outside

lanes

1411 and1412—and thus a region of irrelevancy (for vehicles)3087 is defined outside the lanes.

FIG. 56 illustrates two signatures of different resolutions—one signature (6027′) including five indexes and a second signature (6027″) include less indexes.

The number of indexes may differ from four of five. The first and second signatures may differ from each other by more than a single index.

FIG. 57 illustrates an example of various data structures and processes.

Assuming that the reference information includes multiple (M) cluster structures4974(1)-4974(M). Each cluster structure includes cluster signatures, metadata regarding the cluster signature

For example—first cluster structure4974(1) includes multiple (N1) signatures (referred to as cluster signatures CS) CS(1,1)—CS(1,N1)4975(1,1)-4975(1,N1), and metadata4976(1). Y Yet for another example—M′th cluster structure4974(M) includes multiple (N2) signatures (referred to as cluster signatures CS) CS(M,1)-CS(M,N2)4975(M,1)-4975(M,N2), and metadata4976(M).

It is assumed that there are three cluster structures related to a situation—first cluster structure4974(1), second cluster structure4974(2), and first cluster structure4974(3).

In this scenario only the three cluster structures may be retrieved or processed during a situation related processing.

For example—amedia unit signature4972 may be compared to the three clusters in order to find one or more matching clusters3091.

The object or object that represented by themedia unit signature4972 is detected3092 based on the metadata of the matching concept structure.

FIG. 58 illustrates an example of avehicle3096 that include aprocessor3099, acache3098, and memory unit3097, as well as other components (not shown). The vehicle may communicate with acomputerized system3095 located outside the vehicle.

Vehicle to Vehicle Communication Less Truck Platooning

There may be provided a system and method for V2V communication-less based platooning in which the platooning is not based solely on V2V communication—and in which the following trucks do not blindly follow the leading truck—based solely on commands (conveyed by the V2V communication).

Each truck may determine, even without communicating with other trucks of the convoy, when to engage in the platooning, how do drive during the platooning and if required—how to disengage from the platooning.

Each truck may decide, from time to time—in a continuous or non-continuous manner on a noncontinuous basis whether to maintain in the convoy—the determination can be made when the preceding truck turns into a path that is not a part of the desired path of the truck, when a vehicle cuts between the truck and another member of the convoy, when the preceding truck drives in a dangerous manner).

Each truck may apply at least a partial autonomous driving while being engaged in the platooning—while basing its behavior on sensed information obtained by the truck and by following one or more autonomous driving policy. The sensed information may include image or non-image information, may be sensed by active sensor (LIDAR) or passive sensor, may include, for example information sensed by visual sensor, infrared sensor, thermal sensor, radar, sonar, and the like.

At least partially autonomously man mean performing some driving assistance operation by the vehicle such as at least one out of adaptive cruise control, lane monitoring, maintaining a distance between the truck and a preceding truck of the convoy—till fully autonomous driving. For example—while in the convoy the truck may apply at least one out of adaptive cruise control (keeping safe distance between the truck and preceding truck—for example by measuring the speed of the preceding truck, identify that the brake light of the preceding truck indicate that it stops, identify lane and drivable paths—for example in order to detect the truck from following the preceding truck into the shoulders of the road, track the exact location of the preceding vehicle to maintain alignment between the truck and the preceding truck),

Trucks of the convoy may attempt to exchange information over V2V communication—but the platooning is not based on said communication.

FIG. 59 illustratesmethod5900.

Method

5900 may start by steps S190 and S920.

Step

5910 may include monitoring, by a truck monitor, an environment of the truck to provide monitoring results.Step5910 may include generating sensed information regarding an environment of a truck and processing the sensed information.Step5910 may continue while the truck is driving, before entering the platoon, while platooning, when disengaging from the platoon and after disengaging from the platoon.

Step

5920 may include determining, based on the sensed information, whether to engage in truck platooning with a set of at least one other truck.

Assuming that a certain truck executedmethod5900.

The set of trucks (before or after the certain truck joins them) can be regarded as a convoy.

The decision of whether to engagement may be based on at least some of the following parameters: a size of one or more trucks of the convoy, a size of the certain truck, a shape of one or more trucks of the convoy, a shape of the certain truck, a weight of one or more trucks of the convoy, a weight of the current truck, a path of the convoy, a path of the certain truck, and the like.

The shape, size and weight may affect the fuel saving which is one of the major benefits of the platooning. For example—a heavy truck (for example class7 orclass8 truck) usually will not follow a convoy of light or medium trucks (forexample classes1,2A,2B, or3).

The shape, size and weight may affect yet another parameter that should be taken into account when deciding to engage in platooning. For example—behavior considerations, e.g. time to stop, safe distance, safe speed etc—that differ for different types of trucks.

Step

5920 may also include determining how to engage—including the location within the convoy—whether to become a leading truck, whether to become a following truck, the order of truck in the convoy, and the like.

How to engage may include determined whether the engagement is to be done autonomously or under the control of a human driver.

How to engage may including, timing of engagement, driving the preceded the engagement, and the like. The engagement can be executed while autonomously driving or while the truck is controlled by a driver (human)—with or without driver assistance module involvement.

Step

5920 may be followed bystep5930 of engaging the truck platooning following determination to engage in the truck platooning.

Step

5920 and S930 may include determining a distance between the truck and the preceding truck of the convoy.

The distance may be dependent on the current speed of the convoy, on the risk level associated with the currently passed road, segments, and the like.

Step

5930 may include at least partially autonomously driving the truck based on monitoring results obtained during the engaging, while engaging in the truck platooning.

Step

5930 may also be followed bystep5940 of deciding whether to disengage from the truck platooning.

When determining to disengage then step5940 may be followed bystep5950 of disengaging—leaving the convoy.

Step

5950 may be followed bystep5920—in order to enable the truck to reengage in the truck platooning—either with the previous convoy or with a new convoy.

There may be a predefine delay between consecutive decisions to engage and disengage—in order to reduce multiple repetitions of engagement and disengagements.

Repetitions of disengaging and reengaging may be executed in various manner—for example

- Temporarily disengaging from the platoon when a vehicle that should not belong to the platoon enters a gap between two consecutive trucks of the convoy. Once the vehicle exits said gap the truck may reengage.
- Replacing places between members of the convoy—for example replacing the leading truck—so that the disadvantages resulting from being a leading truck (for example—no fuel saving) can be distributed between different members of the convoy.

The members of the convoy may replace locations within the convoy in a predefined manner, in a pseudorandom manner, or a random manner, following and event (for example after passing a predefined distance, and the like).

Each one of

steps

5920,5930 and5940 may be based in various considerations. These considerations may define, for example, the distance between the certain truck and the preceding truck. The distance may change over time (dependent on other considerations such as speed, risk, and the like).

These considerations may include at least one out of the benefits in platooning (fuel saving, simpler autonomous driving) versus the downfalls of platooning (smaller distances between the trucks).

Any cost function or tradeoff between the benefits and the downfalls of platooning may be applied.

The risk associated with platooning may be determined in various manners and/or may be dependent on one or more factors—such as (a) the state of the road—for example grip level, slipperiness, (b) weather conditions—rain, show, ice, fog, illumination, sunny, windy, twilight, sunrise, (c) manner in which the certain truck is forced to drive, (d) the risk level associated with the environment of the convoy (traffic load, steepness of road segments, velocity of driving, urban versus open country environment), type of vehicles (cars, motorcycles, bicycles, scooters), presence or absence of pedestrians), and the like.

The manner in which the truck was forced to drive while platooning can be evaluated in various manners—for example:

- Whether it followed a safe driving profile (may be determined in advance) while platooning.
- Whether while driving in the convoy the driving of the truck was more aggressive than when driving independently.
- Whether risky events occurred—for example—whether rule traffic violations occurred (passing in red light, driving without stopping in front of a stop sign, exceeding the allowed speed limit).

The risk may be associated in a continuous manner, in a non-continuous manner, randomly, pseudo-randomly, due to events, and the like.

The risk may be based by applying a machine learning process trained to identify risky driving.

The considerations may include the amount of changes in the number of members of the convoy, the rate of location exchanges in the convoy, and the like. For example—too many changes may reduce the safety level of the convoy.

The considerations may also include the physiological state of the driver. For example—if the driver is tired or stressed then joining a convoy as a following truck may allow autonomous driving that may not be allowed without platooning.

The considerations may include the overlap between the current path of the convoy and the desired path of the truck. The method may allow deviations from the desired path—but the deviations should still be more beneficial (for example in fuel saved) than skipping these deviations.

The considerations or the tradeoff may be leant by supervised or unsupervised machine learning process.

The entrance of one or more vehicles between the certain truck and a previous truck of the convoy may cause the certain truck to disengage—either temporarily or not. Such one or more vehicles (that should not belong to the convoy) are usually associated with lower trust level (for example are not autonomously driven).

FIG. 60 illustratesmethod5902.

Method

5902 differ frommethod5900 ofFIG. 59 by including steps S960 and S970.

Step

5960 may include receiving vehicle to vehicle communication, wherein the vehicle to vehicle communication includes commands (for example from a leading truck) for determining driving of trucks while engaged in the truck platooning.

Step

5960 may be followed bystep5970 of responding to or ignoring the commands.

Step

5970 may include at least partially autonomously driving the truck regardless of the commands.

Step

5970 may include analyzing an accuracy of the commands and generating an alerts when the commands are inaccurate.

For example—if the convoy stops and the commands are to accelerate—then the commands are inaccurate—even may be a result of a jamming attempt.

Step

5970 may include (instead of or in addition to the generating of the alert) increasing the risk associated with the platooning and/or leaving the convoy.

Any reference in the specification to a method should be applied mutatis mutandis to a system capable of executing the method and should be applied mutatis mutandis to a non-transitory computer readable medium that stores instructions that once executed by a computer result in the execution of the method.

Any reference in the specification to a system and any other component should be applied mutatis mutandis to a method that may be executed by a system and should be applied mutatis mutandis to a non-transitory computer readable medium that stores instructions that may be executed by the system.

Any reference in the specification to a non-transitory computer readable medium should be applied mutatis mutandis to a system capable of executing the instructions stored in the non-transitory computer readable medium and should be applied mutatis mutandis to method that may be executed by a computer that reads the instructions stored in the non-transitory computer readable medium.

Any combination of any module or unit listed in any of the figures, any part of the specification and/or any claims may be provided. Especially any combination of any claimed feature may be provided.

Any reference to the term “comprising” or “having” should be interpreted also as referring to “consisting” of “essentially consisting of”. For example—a method that comprises certain steps can include additional steps, can be limited to the certain steps or may include additional steps that do not materially affect the basic and novel characteristics of the method—respectively.

The invention may also be implemented in a computer program for running on a computer system, at least including code portions for performing steps of a method according to the invention when run on a programmable apparatus, such as a computer system or enabling a programmable apparatus to perform functions of a device or system according to the invention. The computer program may cause the storage system to allocate disk drives to disk drive groups.

A computer program is a list of instructions such as a particular application program and/or an operating system. The computer program may for instance include one or more of: a subroutine, a function, a procedure, an object method, an object implementation, an executable application, an applet, a servlet, a source code, an object code, a shared library/dynamic load library and/or other sequence of instructions designed for execution on a computer system.

The computer program may be stored internally on a computer program product such as non-transitory computer readable medium. All or some of the computer program may be provided on non-transitory computer readable media permanently, removably or remotely coupled to an information processing system. The non-transitory computer readable media may include, for example and without limitation, any number of the following: magnetic storage media including disk and tape storage media; optical storage media such as compact disk media (e.g., CD-ROM, CD-R, etc.) and digital video disk storage media; nonvolatile memory storage media including semiconductor-based memory units such as FLASH memory, EEPROM, EPROM, ROM; ferromagnetic digital memories; MRAM; volatile storage media including registers, buffers or caches, main memory, RAM, etc. A computer process typically includes an executing (running) program or portion of a program, current program values and state information, and the resources used by the operating system to manage the execution of the process. An operating system (OS) is the software that manages the sharing of the resources of a computer and provides programmers with an interface used to access those resources. An operating system processes system data and user input, and responds by allocating and managing tasks and internal system resources as a service to users and programs of the system. The computer system may for instance include at least one processing unit, associated memory and a number of input/output (I/O) devices. When executing the computer program, the computer system processes information according to the computer program and produces resultant output information via I/O devices.

In the foregoing specification, the invention has been described with reference to specific examples of embodiments of the invention. It will, however, be evident that various modifications and changes may be made therein without departing from the broader spirit and scope of the invention as set forth in the appended claims.

Moreover, the terms “front,” “back,” “top,” “bottom,” “over,” “under” and the like in the description and in the claims, if any, are used for descriptive purposes and not necessarily for describing permanent relative positions. It is understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the invention described herein are, for example, capable of operation in other orientations than those illustrated or otherwise described herein.

Those skilled in the art will recognize that the boundaries between logic blocks are merely illustrative and that alternative embodiments may merge logic blocks or circuit elements or impose an alternate decomposition of functionality upon various logic blocks or circuit elements. Thus, it is to be understood that the architectures depicted herein are merely exemplary, and that in fact many other architectures may be implemented which achieve the same functionality.

Any arrangement of components to achieve the same functionality is effectively “associated” such that the desired functionality is achieved. Hence, any two components herein combined to achieve a particular functionality may be seen as “associated with” each other such that the desired functionality is achieved, irrespective of architectures or intermedial components. Likewise, any two components so associated can also be viewed as being “operably connected,” or “operably coupled,” to each other to achieve the desired functionality.

Furthermore, those skilled in the art will recognize that boundaries between the above described operations merely illustrative. The multiple operations may be combined into a single operation, a single operation may be distributed in additional operations and operations may be executed at least partially overlapping in time. Moreover, alternative embodiments may include multiple instances of a particular operation, and the order of operations may be altered in various other embodiments. Also for example, in one embodiment, the illustrated examples may be implemented as circuitry located on a single integrated circuit or within a same device. Alternatively, the examples may be implemented as any number of separate integrated circuits or separate devices interconnected with each other in a suitable manner.

Also for example, the examples, or portions thereof, may implemented as soft or code representations of physical circuitry or of logical representations convertible into physical circuitry, such as in a hardware description language of any appropriate type.

Also, the invention is not limited to physical devices or units implemented in non-programmable hardware but can also be applied in programmable devices or units able to perform the desired device functions by operating in accordance with suitable program code, such as mainframes, minicomputers, servers, workstations, personal computers, notepads, personal digital assistants, electronic games, automotive and other embedded systems, cell phones and various other wireless devices, commonly denoted in this application as ‘computer systems’.

However, other modifications, variations and alternatives are also possible. The specifications and drawings are, accordingly, to be regarded in an illustrative rather than in a restrictive sense.

In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word ‘comprising’ does not exclude the presence of other elements or steps then those listed in a claim. Furthermore, the terms “a” or “an,” as used herein, are defined as one or more than one. Also, the use of introductory phrases such as “at least one” and “one or more” in the claims should not be construed to imply that the introduction of another claim element by the indefinite articles “a” or “an” limits any particular claim containing such introduced claim element to inventions containing only one such element, even when the same claim includes the introductory phrases “one or more” or “at least one” and indefinite articles such as “a” or “an.” The same holds true for the use of definite articles. Unless stated otherwise, terms such as “first” and “second” are used to arbitrarily distinguish between the elements such terms describe. Thus, these terms are not necessarily intended to indicate temporal or other prioritization of such elements. The mere fact that certain measures are recited in mutually different claims does not indicate that a combination of these measures cannot be used to advantage.

While certain features of the invention have been illustrated and described herein, many modifications, substitutions, changes, and equivalents will now occur to those of ordinary skill in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.