Summary of the invention
The embodiment of the present application technical problem to be solved is, provides a kind of driving behavior model based on SVM algorithmMethod for building up and device, accurate driving behavior model can be trained, lane-change is effectively predicted according to the driving behavior model and is drivenThe safety for sailing behavior improves the safety of driving.
In order to solve the above-mentioned technical problem, the foundation of this application provides a kind of driving behavior model based on SVM algorithmMethod, comprising: data are observed in the driving obtained in forecast interval, and forecast interval is divided into K run duration and is held high, when to KBetween driving observation data in window carry out feature extraction and obtain K feature vector and be weighted summation obtaining weighted feature vector,The weighted feature vector is trained to obtain SVM prediction model according to SVM algorithm.
Implement the embodiment of the present application, be divided into weighted sum by carrying out to forecast interval, in utilization SVM algorithm to weightingFeature vector is trained to obtain SVM prediction model, and the later use SVM prediction model is classified to observation data, can be mentionedThe accuracy that high driving behavior classification differentiates, improves the safety of driving.
In a kind of possible design, feature vector includes the mean value and variance of running velocity.
In a kind of possible design, it includes: according to preset time window that the forecast interval, which is divided into K time window,The forecast interval is divided into K runing time window by length or time window interval, wherein the equal length of each time window.
In a kind of possible design, K time window is SW1,SW2,…,SWK, the corresponding feature of K time window toAmount is x1,x2,…,xK, the corresponding weight of K feature vector is w1,w2,…,wK, wherein w1<w2<…<wK.I.e. the time gets overThe corresponding weight of time window rearward is bigger, on the contrary the time it is more forward the corresponding weight of time window it is smaller.
In a kind of possible design, feature vector xkWeight wkShown using following formula game clock:
Wherein, C is weight reduction coefficient, 0≤C < 1, dk=(1-C)K-k, k indicate time window serial number, k=1,2 ...,K, and k is the integer greater than 0.
In a kind of possible design, time window length, time window interval and weight reduction coefficient are inputted into SVM classifierThe weighted feature vector is classified to obtain driving behavior classification, such as: when the value that input SVM classifier obtains is+1,Expression driving behavior classification is dangerous driving classification, and when obtained value is -1, expression driving behavior classification is normal driving classification.
Second aspect establishes device this application provides a kind of driving behavior model, comprising:
Acquiring unit, for obtaining the observation data of the driving in forecast interval；
Division unit, for forecast interval to be divided into K time window；
Extraction unit obtains K feature vector for carrying out feature extraction to the driving observation data in K time window；
Weighted units obtain weighted feature vector for being weighted summation to the K feature vector；
Training unit, for being trained to obtain SVM prediction model to the weighted feature vector according to SVM algorithm.
In a kind of possible design, described eigenvector includes the mean value and variance of running velocity.
In a kind of possible design, the division unit is specifically used for:
The forecast interval is divided into k time window according to preset time window length and time window interval.
In a kind of possible design, K time window is SW1,SW2,…,SWK, K time window corresponding K specialSign vector is x1,x2,…,xK, the corresponding K weight of K feature vector is w1,w2,…,wK, wherein w1<w2<…<wK。
In a kind of possible design, feature vector xkWeight wkShown using following formula game clock:
Wherein, C is weight reduction coefficient, 0≤C < 1, dk=(1-C)K-k, k indicate time window serial number, k=1,2 ...,K, and k is the integer greater than 0.
The another aspect of the application provides a kind of device, and the foundation of the driving behavior model of above-mentioned first aspect may be implementedMethod.Such as described device can be chip (such as baseband chip or communication chip etc.) or processing server.It can be by softPart, hardware or the corresponding software realization above method is executed by hardware.
It in one possible implementation, include processor, memory in the structure of described device；The processor quiltIt is configured to that described device is supported to execute corresponding function in above-mentioned communication means.Memory is saved for coupling with processorThe necessary program of described device (instruction) and/or data.Optionally, the communication device can also include communication interface for branchHold the communication between described device and other network elements.
In alternatively possible implementation, described device may include the list for executing corresponding actions in the above methodElement module.
In another possible implementation, including processor and R-T unit, the processor and the transmitting-receiving fillSet coupling, the processor for executing computer program or instruction, with control the R-T unit carry out information reception andIt sends；When the processor executes the computer program or instruction, the processor is also used to realize the above method.ItsIn, the R-T unit can be transceiver, transmission circuit or input/output interface.When the communication device is chip, instituteStating R-T unit is transmission circuit or input/output interface.
When described device is chip, transmission unit can be output unit, such as output circuit or communication interface；It connectsReceiving unit can be input unit, such as input circuit or communication interface.When the communication device is the network equipment, sendUnit can be transmitter or transmitter；Receiving unit can be receiver or receiver.
The another aspect of the application provides a kind of device, which includes: memory and processor；Wherein, the storageBatch processing code is stored in device, and the processor executes each side for calling the program code stored in the memoryMethod described in face.
The another aspect of the application has been mentioned for a kind of computer readable storage medium, in the computer readable storage mediumIt is stored with instruction, when run on a computer, so that computer executes method described in above-mentioned various aspects.
The another aspect of the application provides a kind of computer program product comprising instruction, when it runs on computersWhen, so that computer executes method described in above-mentioned various aspects.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understoodThe application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, andIt is not used in restriction the application.Meanwhile in the description of the present application, term " first ", " second " etc. are only used for distinguishing description, withoutIt can be interpreted as indication or suggestion relative importance.It will be clear to one skilled in the art that these details itsThe application also may be implemented in its embodiment.In other situations, it omits to well-known system, device, circuit and sideThe detailed description of method, so as not to obscure the description of the present application with unnecessary details.
Support vector machines (Support Vector Machine, SVM) algorithm: SVM algorithm is to develop the mid-90A kind of machine learning method based on Statistical Learning Theory come, it is extensive to improve learning machine by seeking structuring least riskAbility realizes that the minimum of empiric risk and fiducial range can also obtain to reach in the case where statistical sample amount is lessThe purpose of good statistical law.It is a kind of two classification model, and basic model is defined as the maximum of the interval on feature spaceLinear classifier, i.e. the learning strategy of support vector machines is margin maximization, can finally be converted into a convex quadratic programmingThe solution of problem.
In order to illustrate technical solution described herein, the following is a description of specific embodiments.
It is a kind of foundation side of driving behavior model based on SVM algorithm provided by the embodiments of the present application referring to Fig. 1, Fig. 1The flow diagram of method, in the embodiment of the present application, which comprises
Data are observed in S101, the driving obtained in forecast interval.
Specifically, forecast interval is a period, the initial time of forecast interval is the company that preset duration occurs for vehicleAt the time of continuous lateral displacement, at the time of the finish time of forecast interval is that vehicle is more than adjacent lane line.Drive observation dataFor time series data, that is, driving observation data includes acquisition moment and the acquisition moment collected data.Terminal device canDriving observation data are acquired in forecast interval according to preset collection period.
S102, forecast interval is divided into K runing time window.
Specifically, runing time window is a period, forecast interval is divided into K runing time window, when K operationBetween the length of window can be equal, the length of each runing time window, which can according to need, to be configured, and the embodiment of the present application is not madeLimitation.The length of K runing time window can also be unequal, such as: the length of K runing time window is according to default stepIncreasing or decreasing.K time window can be in continuously distributed, can also be in discontinuous distribution, the embodiment of the present application is with no restriction.
S103, K feature vector is obtained to the driving observation data progress feature extraction in K time window.
Specifically, the parameter in feature vector include lateral velocity difference average value between adjacent two vehicle, adjacent two vehicle itBetween longitudinal velocity difference average value, the lateral velocity difference variance between adjacent two vehicle, longitudinal velocity difference variance, the phase of adjacent two vehicleThe centroid distance variance between centroid distance average value, adjacent two vehicle between adjacent two vehicles.
S104, it is weighted summation to K feature vector obtains weighted feature vector.
Specifically, the corresponding weight of each vehicle characteristics vector, according to the weight of each vehicle characteristics vector to K spySign vector is weighted summation and obtains weighted feature vector.
S105, the weighted feature vector is trained to obtain SVM prediction model according to SVM algorithm.
Specifically, being trained to obtain SVM prediction model (also referred to as svm classifier to weighted feature vector according to SVM algorithmDevice), it is subsequent that the observation data in vehicle operation are classified using SVM classifier to obtain driving behavior classification, such as:Collected observation data a value will be exported after being input to SVM classifier in forecast interval, and when which is+1, indicate predictionDriving behavior in section is dangerous driving classification, indicates that the driving behavior in forecast interval is normal driving row when which is -1For.
Implement the embodiment of the present application, be divided into weighted sum by carrying out to forecast interval, in utilization SVM algorithm to weightingFeature vector is trained to obtain SVM prediction model, and the later use SVM prediction model is classified to observation data, can be mentionedThe accuracy that high driving behavior classification differentiates, improves the safety of driving.
A referring to fig. 2 is a kind of the another of method for building up of the driving behavior model based on SVM provided by the embodiments of the present applicationOne flow diagram, in the embodiment of the present application, which comprises
S201, configurator timing window length and time window interval.
Specifically, time window length indicates the duration of time window, two adjacent time windows of time window time intervalBetween time interval, time window length and time window interval are for being divided into multiple time windows to forecast interval.
Data are observed in S202, the driving obtained in forecast interval.
Specifically, forecast interval is a period, the initial time of forecast interval is the company that preset duration occurs for vehicleAt the time of continuous lateral displacement, at the time of finish time is that vehicle is more than adjacent lane line.Such as: referring to fig. 2 shown in b, predictionThe initial time in section is t0, the finish time of forecast interval is tw。
S203, forecast interval is divided by K time window according to time window length and time window interval.
Specifically, by forecast interval, temporally window length and time window interval are divided into the different time windows of K, at KBetween the length of window can be equal, K time window can also be able to be in discontinuous distribution in continuously distributed.For example, with reference to Fig. 2 c instituteShow, forecast interval is [t0,tw], forecast interval is divided by K different time windows according to time window length and time window intervalSW1、 SW2、…、SWK。
S204, K feature vector is obtained to the driving observation data progress feature extraction in K time window.Specifically, rightDriving observation data (for a kind of time series data) in each time window carries out feature extraction and (extracts SWkInterior vehicle operationFeature vector x of the mean value and variance of speed as time windowk), to obtain the feature vector { x of K d dimension1,x2,...,xk}.In the embodiment of the present application, the parameter in feature vector includes lateral velocity difference average value between two adjacent vehicles, phaseThe lateral velocity difference variance between longitudinal velocity difference average value, adjacent two vehicle, the longitudinal velocity of adjacent two vehicle between adjacent two vehiclesThe centroid distance variance between centroid distance average value, adjacent two vehicle between poor variance, adjacent two vehicle.
S205, it summation is weighted to K vehicle operation characteristic vector obtains weighted feature vector.
Specifically, obtaining the vehicle operation characteristic vector of entire forecast interval using Exchanger Efficiency with Weight Coefficient MethodItsMiddle wkFor the weight of time window SWk.
Wherein, close to the finish time t of forecast intervalwTime window compared with time window before have bigger weight(risk that newest motion state data can more reflect driving behavior), therefore the exponential type based on time gap is used to be reduced systemNumber function dkTo calculate the weight w of each time windowk:
Wherein, C is weight reduction coefficient, 0≤C < 1, dk=(1-C)K-k, when k is indicatedBetween window serial number, k=1,2 ..., K, and k is integer greater than 0.Such as: d1=(1-C)K-1≤1,dk=1, therefore dkMeaning beDistance Time window SWk is remoter, then the value that characteristic value is reduced in the time window is bigger, the shared weight w in all time windowskIt is smaller.The weight for being equal to each time window as C=0 is equal.
S206, the weighted feature vector is trained to obtain SVM prediction model according to SVM algorithm.
Specifically, xi∈RdFeature vector for the one group of d dimension obtained in forecast interval by Data Dimensionality Reduction, the feature of d dimensionThe feature vector of vector expression vehicle operation.Given N group driving sample data { x to be markedi, yi, i=1,2 ..., N,Following discriminant function D (z) is established by learning training to determine to drive classification belonging to observation data z:
With every group of vehicle operation characteristic vector xiCorresponding discriminant value passes throughyi∈ {+1, -1 } indicates that+1 indicates dangerous driving behavior, and -1 indicates normal driving behavior；K(xi,yi) it is kernel function, i.e., willOriginal vector xiAnd yiMake inner product after projecting to feature space, B is deviation factors item, and α is coefficient vector, and the value of α meets constraintIt is required thatAnd W (α) is made to reach maximum value simultaneously,The precision of SVM classifier is main and kernel function K (xi,yi) selection it is related with feature vector x.
The key parameter of SVM prediction model includes: time window length, time window interval and weight reduction coefficient C.Terminal is setIt is standby to determine the parameters value in SVM prediction model using ten folding interior extrapolation methods.
Wherein, it in the case where ensureing that prediction result FP (False Positive) is lower than 5%, is obtained using grid data serviceIt must make maximum preceding 10 combining parameter values of TP rate value.
| Serial number | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| TP/% | 89.4 | 89.1 | 88.9 | 88.5 | 88.2 | 87.9 | 87.5 | 87.0 | 87.5 | 87.2 |
| S/s | 0.5 | 0.4 | 0.4 | 0.5 | 0.6 | 0.7 | 0.7 | 0.6 | 0.7 | 0.7 |
| I/s | 0.3 | 0.2 | 0.3 | 0.4 | 0.3 | 0.2 | 0.3 | 0.4 | 0.4 | 0.5 |
| C | 0.1 | 0.15 | 0.1 | 0.05 | 0.1 | 0.15 | 0.1 | 0.05 | 0.05 | 0 |
Table 1
In table 1, TP (true positive) indicates accuracy rate, S: time window length, and I indicates time window interval, and C is indicatedWeight reduction coefficient.
Then, weighted feature vector is trained according to SVM algorithm to obtain SVM prediction model (also referred to as SVM classifier),It is subsequent that the observation data in vehicle operation are classified using SVM classifier to obtain driving behavior classification, such as: it will beCollected observation data export a value after being input to SVM classifier in forecast interval, when which is+1, indicate forecast intervalInterior driving behavior is dangerous driving classification, indicates that the driving behavior in forecast interval is normal driving behavior when which is -1.
Wherein, in the embodiment of the present application, preferable in order to reach when time window length and time window interval increase simultaneouslyPrediction effect (i.e. higher TP value), weight reduction coefficient is generally intended to 0.Should the result shows that, when each time windowWhen length increases and the section not being overlapped between time window increases, i.e., when the quantity K of the time window divided in forecast interval is reduced,Reduce and the difference (weight that C=0 represents each time window is equal) of prediction contribution proportion is beneficial to obtain between time windowHigher predictablity rate.Meanwhile the case where unallocated time window (K=1, and the section of time window is [t0, tw]) carry out pairThan test, it predicts that TP value is only 86.3% as the result is shown, and the feature based on time window for showing that the embodiment of the present application proposes mentionsMethod is taken to be conducive to improve the precision of prediction of SVM.Finally, best parameter group 1 predicts test sample in selection table, in advanceTP the and FP value for surveying result is respectively 88.7% and 4.0%, substantially meets the requirement of dangerous lane-change early warning.
Above-mentioned Fig. 2 illustrates a kind of foundation side of driving behavior model based on SVM algorithm of the embodiment of the present applicationMethod.
Fig. 3 is referred to, Fig. 3 is a kind of structural schematic diagram of device provided by the embodiments of the present application, which may includeAcquiring unit 301, division unit 302, extraction unit 303, weighted units 304 and training unit 305.
Acquiring unit 301, for obtaining the observation data of the driving in forecast interval.
Division unit 302, for forecast interval to be divided into K time window；K is the integer greater than 1.
Extraction unit 303, for in K time window driving observation data carry out feature extraction obtain K feature toAmount.
Weighted units 304 obtain weighted feature vector for being weighted summation to the K feature vector.
Training unit 305, for being trained to obtain SVM prediction model to the weighted feature vector according to SVM algorithm.
In a kind of possible embodiment, the parameter in described eigenvector includes the laterally speed between two adjacent vehiclesSpend the longitudinal velocity difference average value between poor average value, adjacent two vehicle, the lateral velocity difference variance between adjacent two vehicle, adjacent twoThe centroid distance variance between centroid distance average value, adjacent two vehicle between the longitudinal velocity difference variance of vehicle, adjacent two vehicle.
Summarize in a kind of possible embodiment, division unit 302 is specifically used for:
The forecast interval is divided into K time window according to preset time window length and time window interval.
In a kind of possible embodiment, the K time window is SW1,SW2,…,SWK, the K time window is respectivelyCorresponding feature vector is x1,x2,…,xK, the corresponding weight of K feature vector is w1,w2,…,wK, wherein w1<w2<…<wK。
In a kind of possible embodiment, feature vector xkWeight wkIt is indicated using following formula:
Wherein, C is weight reduction coefficient, 0≤C < 1, dk=(1-C)K-k, k indicate time window serial number, k=1,2 ...,K, and k is the integer greater than 0.
Device 3 can be terminal device, described device 3 or the field programmable gate array for realizing correlation function(field-programmable gate array, FPGA), special integrated chip, System on Chip/SoC (system on chip,SoC), central processing unit (central processor unit, CPU), network processing unit (network processor, NP),Digital signal processing circuit, microcontroller (micro controller unit, MCU), can also use programmable controller(programmable logic device, PLD) or other integrated chips.
The embodiment of the present application and the embodiment of the method for Fig. 2 a are based on same design, and bring technical effect is also identical, toolBody process can refer to the description of the embodiment of the method for Fig. 2 a, and details are not described herein again.
Fig. 4 is a kind of apparatus structure schematic diagram provided by the embodiments of the present application, and hereinafter referred to as device 4, device 4 can integrateIn terminal device above-mentioned, as shown in figure 4, the device includes: memory 402, processor 401 and transceiver 403.
Memory 402 can be independent physical unit, can be connect by bus with processor 401, transceiver 403.Memory 402, processor 401, transceiver 403 also can integrate together, pass through hardware realization etc..
Memory 402 is used to store the program for realizing above method embodiment or Installation practice modules, processingDevice 401 calls the program, executes the operation of above method embodiment.
Optionally, pass through when some or all of in the driving behavior classification method based on SVM algorithm of above-described embodimentWhen software realization, device can also only include processor.Memory for storing program is located at except device, and processor passes throughCircuit/electric wire is connect with memory, for reading and executing the program stored in memory.
Processor can be central processing unit (central processing unit, CPU), network processing unitThe combination of (network processor, NP) or CPU and NP.
Processor can further include hardware chip.Above-mentioned hardware chip can be specific integrated circuit(application-specific integrated circuit, ASIC), programmable logic device (programmableLogic device, PLD) or combinations thereof.Above-mentioned PLD can be Complex Programmable Logic Devices (complexProgrammable logic device, CPLD), field programmable gate array (field-programmable gateArray, FPGA), Universal Array Logic (generic array logic, GAL) or any combination thereof.
Memory may include volatile memory (volatile memory), such as random access memory(random-access memory, RAM)；Memory also may include nonvolatile memory (non-volatile), such as flash memory (flash memory), hard disk (hard disk drive, HDD) or solid state hard disk memory(solid-state drive, SSD)；Memory can also include the combination of the memory of mentioned kind.
In above-described embodiment, transmission unit or transmitter execute the step of above-mentioned each embodiment of the method is sent, and receive singleMember or receiver execute the step of above-mentioned each embodiment of the method receives, and other steps are executed by other modules or processor.HairSend unit and receiving unit that can form Transmit-Receive Unit, receiver and transmitter can form transceiver.
The embodiment of the present application also provides a kind of computer storage mediums, are stored with computer program, the computer programFor executing the method for building up of the driving behavior model provided by the above embodiment based on SVM algorithm.
The embodiment of the present application also provides a kind of computer program products comprising instruction, when it runs on computersWhen, so that computer executes the method for building up of the driving behavior model provided by the above embodiment based on SVM algorithm.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer programProduct.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the applicationApply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, which can be used in one or more,The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) producesThe form of product.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present applicationFigure and/or block diagram describe.It should be understood that can be realized by computer program instructions each in flowchart and/or the block diagramThe combination of process and/or box in process and/or box and flowchart and/or the block diagram.It can provide these computersProcessor of the program instruction to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devicesTo generate a machine, so that generating use by the instruction that computer or the processor of other programmable data processing devices executeIn the dress for realizing the function of specifying in one or more flows of the flowchart and/or one or more blocks of the block diagramIt sets.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spyDetermine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram orThe function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that countingSeries of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer orThe instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram oneThe step of function of being specified in a box or multiple boxes.