Disclosure of Invention
The application provides a map updating method, a map updating device, electronic equipment and a computer readable storage medium, which can solve the problem of low map updating accuracy.
In a first aspect of the present application, there is provided a map updating method, the method comprising:
Acquiring a current image and a historical image of a map corresponding to a region to be detected;
determining distinguishing characteristic information between the current image and the historical image based on the trained twin network;
determining at least one change positioning frame of the historical image relative to the current image based on the distinguishing characteristic information;
The map is updated based on the at least one changed positioning frame.
In one possible implementation manner, the determining the distinguishing characteristic information between the current image and the historical image includes:
extracting features of the current image to obtain a first feature point set;
extracting features of the historical images to obtain a second feature point set;
determining a set of distinguishing points between the first set of feature points and the second set of feature points;
based on the set of distinguishing points, distinguishing feature information is generated.
In one possible implementation manner, the determining at least one change positioning frame of the historical image relative to the current image based on the distinguishing characteristic information includes:
determining at least one target feature point based on the distinguishing feature information;
and determining at least one change positioning frame with a preset shape based on the target characteristic points.
In another possible implementation manner, determining at least one target feature point based on the distinguishing feature information includes:
upsampling the distinguishing characteristic information to obtain fusion characteristic information;
And acquiring at least one target feature point based on the fusion feature information.
In another possible implementation manner, the shape of the variable positioning frame is rectangular, and the determining at least one preset shape variable positioning frame based on the target feature point comprises:
respectively determining the category and margin data corresponding to each target feature point;
And respectively determining rectangular change positioning frames corresponding to each target feature point based on the category and the margin data.
In yet another possible implementation manner, updating the map based on the at least one change positioning frame includes:
Determining a change profile based on each change positioning frame;
receiving correction information, correcting the change profile based on the correction information, and generating ground feature profile data;
And updating the map according to the ground feature profile data.
In yet another possible implementation, the twin network is trained by:
sample images of the same sample area in different periods are acquired, and the sample images are provided with corresponding sample positioning frames;
inputting the sample image into an initial twin network to obtain a real-time positioning frame tag output by the initial twin network;
Determining a multi-branch loss function based on the sample positioning frame and the positioning frame tag;
And adjusting parameters of the initial twin network based on the multi-branch loss function to obtain the twin network.
In a second aspect of the present application, there is provided a map updating apparatus comprising:
The acquisition module is used for acquiring a current image and a historical image of a map corresponding to the region to be detected;
the extraction module is used for determining distinguishing characteristic information between the current image and the historical image based on the trained twin network;
A determining module for determining at least one change positioning frame of the history image relative to the current image based on the distinguishing characteristic information;
and the updating module is used for updating the map based on the at least one change positioning frame.
In one possible implementation manner, the extracting module is specifically configured to:
extracting features of the current image to obtain a first feature point set;
extracting features of the historical images to obtain a second feature point set;
determining a set of distinguishing points between the first set of feature points and the second set of feature points;
based on the set of distinguishing points, distinguishing feature information is generated.
In one possible implementation manner, the determining module specifically includes:
a first determining unit configured to determine at least one target feature point based on the distinguishing feature information;
and the second determining unit is used for determining at least one change positioning frame with a preset shape based on the target characteristic points.
In another possible implementation manner, the first determining unit is further configured to:
upsampling the distinguishing characteristic information to obtain fusion characteristic information;
And acquiring at least one target feature point based on the fusion feature information.
In another possible implementation manner, the shape of the above-mentioned variable positioning frame may be rectangular, and the second determining unit is further configured to:
respectively determining the category and margin data corresponding to each target feature point;
And respectively determining rectangular change positioning frames corresponding to each target feature point based on the category and the margin data.
In yet another possible implementation manner, the updating module is further configured to:
Determining a change profile based on each change positioning frame;
receiving correction information, correcting the change profile based on the correction information, and generating ground feature profile data;
And updating the map according to the ground feature profile data.
In a further possible implementation manner, the device further comprises a training module, specifically configured to:
sample images of the same sample area in different periods are acquired, and the sample images are provided with corresponding sample positioning frames;
inputting the sample image into an initial twin network to obtain a real-time positioning frame tag output by the initial twin network;
Determining a multi-branch loss function based on the sample positioning frame and the positioning frame tag;
And adjusting parameters of the initial twin network based on the multi-branch loss function to obtain the twin network.
In a third aspect of the present application, there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the map updating method according to the first aspect of the present application when executing the program.
In a fourth aspect of the present application, there is provided a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the map updating method of the first aspect of the present application.
The technical scheme provided by the application has the beneficial effects that:
According to the application, the distinguishing characteristic information between the current image and the historical image is determined through the twin network, so that the twin network has strong anti-interference capability, the obtained distinguishing characteristic information has better characteristic characterization capability, and the change positioning frame is determined based on the distinguishing characteristic information, so that the map is updated, the false detection of map change can be effectively reduced, and the accuracy of map updating is improved.
In addition, the change identification of the region is realized based on the change positioning frame, the map change region can be locked quickly, compared with the change identification of the pixel level of semantic segmentation in the prior art, the update efficiency of map data is greatly improved, the requirements of a network model on image resolution and image quality are reduced, and the accuracy of the change positioning of the map data is ensured.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the application.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein includes all or any element and all combination of one or more of the associated listed items.
For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application will be described in further detail with reference to the accompanying drawings.
The map is a graph or image which selectively represents several phenomena of earth (or other stars) on plane or sphere in two-dimensional or multidimensional form and means according to a certain rule, has strict mathematical foundation, symbol system and character annotation, and can scientifically reflect the distribution characteristics of natural and socioeconomic phenomena and the interrelation thereof by using map summarizing principle.
Map updating is a work of correcting map contents to improve the accuracy and maintain map behavior according to the actual state of change of the area shown in the map. The shorter the updating period is, the stronger the map is, the higher the value is, and the updating method has two kinds of comprehensive retesting (editing) and local repair (editing).
The map updating method provided by the application can realize efficient and accurate map updating, thereby ensuring the freshness of the map data.
Artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) is the theory, method, technique, and application system that simulates, extends, and extends human intelligence using a digital computer or a machine controlled by a digital computer, perceives the environment, obtains knowledge, and uses the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.
The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.
Computer Vision (CV) is a science of how to "look" at a machine, and more specifically, to replace a camera and a Computer to perform machine Vision such as identifying and measuring a target by human eyes, and further perform graphic processing, so that the Computer is processed into an image more suitable for human eyes to observe or transmit to an instrument to detect. As a scientific discipline, computer vision research-related theory and technology has attempted to build artificial intelligence systems that can acquire information from images or multidimensional data. Computer vision techniques typically include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content/behavior recognition, three-dimensional object reconstruction, 3D techniques, virtual reality, augmented reality, synchronous positioning, and map construction, among others, as well as common biometric recognition techniques such as face recognition, fingerprint recognition, and others.
The scheme provided by the embodiment of the application relates to an artificial intelligence map updating technology, and is specifically described by the following embodiment.
The key point of map updating is to perform change detection on images of the same region corresponding to different periods of the map, the conventional change detection algorithm based on deep learning mostly utilizes different image deep learning characteristics to generate differential images or adopts a strategy of learning change relation among pixel blocks to complete the change detection, the change detection is a typical algorithm based on semantic segmentation, different types are given to changed pixels and unchanged pixels, and the following problems exist in the detection of a changed building by taking a U-net+FPN+ basic model mode which is conventional in a semantic segmentation scheme in the prior art as an example:
(1) The false detection of pixel level classification is too many, the integrity is lacking, and the manual workload is increased. The end-to-end semantic segmentation algorithm can accurately predict small changes of pixel levels, is sensitive to color light, and causes false detection of a large number of changes, and the accuracy is greatly reduced although the recall rate is ensured.
(2) The inability to filter individual minor architectural changes that are not emphasized results in a lot of wasted information during the inspection process. Such as individual house refurbishment, which often occurs in many large villages, this level of change is not necessarily detected, and does not contribute to the background data.
(3) The semantic segmentation scheme is based on pixel-level detection, and cannot complete instance-level building detection. The pixel block can only extract local features and cannot be directly applied to data fusion, meanwhile, the context of the image is not fully utilized aiming at the classification of the pixel level, the features of different areas of the image cannot be shared, and the classification performance is limited.
(4) The cost of data annotation is too high. The semantic segmentation scheme needs a satellite image with high quality as training data, and each changed building is marked based on pixels, so that the observation and operation costs are too high.
The map updating method provided by the application is based on a model of deep learning target detection, combines a twin network structure to extract change characteristics, and adopts end-to-end to finish map change area detection so as to realize efficient and accurate map updating.
The application provides a map updating method, a map updating device, electronic equipment and a computer readable storage medium, and aims to solve the technical problems in the prior art.
The following describes the technical scheme of the present application and how the technical scheme of the present application solves the above technical problems in detail with specific embodiments. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.
As shown in fig. 1, the map updating method of the present application may be applied to the scene shown in fig. 1, specifically, after the server 103 receives the current image 101 and the history image 102 of the map corresponding to the to-be-detected area sent by the terminal, the distinguishing feature information between the current image 101 and the history image 102 is identified based on the twin network 104, and at least one change positioning frame 105 of the history image 102 relative to the current image 101 is determined, so as to update the map based on the at least one change positioning frame 105, and obtain an updated map 106.
In the scenario shown in fig. 1, the map updating method may be performed in the server, or in other scenarios, may be performed in the terminal.
It will be appreciated by those skilled in the art that the "terminal" as used herein may be a Mobile phone, a tablet computer, a PDA (Personal digital assistant), a MID (Mobile INTERNET DEVICE), etc., and the "server" may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.
In an embodiment of the present application, as shown in fig. 2, a method for updating a map is provided, where the method may include the following steps:
s201, acquiring a current image and a historical image of a map corresponding to a region to be detected.
In the embodiment of the application, after the area to be detected which needs to be updated is determined, the map data corresponding to the area to be detected is extracted so as to acquire the map data of the same area in different periods, namely the current image and the historical image of the map corresponding to the area to be detected.
The current image or the historical image can be obtained through the existing GIS (Geographic Information System ) data, and can also be directly obtained based on remote sensing data.
S202, distinguishing characteristic information between the current image and the historical image is determined based on the trained twin network.
The twin neural network (Siamese neural network), also called twin neural network, is a coupling framework established based on two artificial neural networks, takes two samples as input, and outputs the characterization of embedding the two samples into a high-dimensional space so as to compare the similarity degree of the two samples.
Specifically, the detection network may be based on a target detection algorithm such as FCOS (Fully Convolutional One-Stage Object Detection, full-convolution single-stage detection) or SSD (Single Shot MultiBox Detector, single-stage multi-frame target detection) or fast R-CNN (fast Region-CNN, fast Region detection based on convolutional neural network).
S203, determining at least one change positioning frame of the historical image relative to the current image based on the distinguishing characteristic information.
The change positioning frame characterizes a change condition of a region change area of the historical image relative to the current image, namely, the change condition on the same area position in the map, wherein the change can comprise building of a building, transformation of a road, building of a bridge and the like.
The shape of the variable positioning frame may be rectangular, circular, or polygonal generated based on the profile of the variable object, which is not particularly limited in this embodiment.
And S204, updating the map based on at least one change positioning frame.
In the embodiment, the distinguishing characteristic information between the current image and the historical image is determined through the twin network, so that the twin network has strong anti-interference capability, the obtained distinguishing characteristic information has better characteristic characterization capability, and the change positioning frame is determined based on the distinguishing characteristic information, so that the map is updated, the false detection of map change can be effectively reduced, and the accuracy of map updating is improved.
Next, a map updating method provided in this embodiment will be specifically described by taking a twin network based on the FCOS algorithm as an example. As shown in fig. 3, the model architecture of the FCOS algorithm-based twin network may include a Backbone (Backbone network) module, a FPN (Feature Pyramid Network ) module, and a multi-branch loss function module.
The backup module is used for front-end extraction of image information to generate feature map (feature map) for later network use, and the backup module can generally adopt Resnet or VGGNet structure. The FPN module is used for feature fusion up-sampling and can perform feature processing aiming at multi-scale characteristics of targets in the image. The multi-branch loss function module is used for predicting the change positioning frame and can accurately lock the change area of the map.
In one possible implementation manner provided in the embodiment of the present application, as shown in fig. 4, determining the distinguishing characteristic information between the current image and the historical image in step S202 may include:
(1) Extracting features of the current image to obtain a first feature point set;
(2) Extracting features of the historical images to obtain a second feature point set;
(3) Determining a set of distinguishing points between the first set of feature points and the second set of feature points;
(4) Based on the set of distinguishing points, distinguishing feature information is generated.
In this embodiment, the current image and the comparison image are used together as the input of the twin network, and the feature parameters are extracted based on the backhaul module, where the backhaul module may adopt a Resnet101 structure.
Specifically, the current image and the comparison image share Resnet structures respectively extract corresponding features, the downsampling of the resnet structure network is divided into 5 stages, namely C1, C2, C3, C4 and C5, and the feature map output by each stage is different in scale. Because the semantic information of the C1 and the C2 stages is too little, the feature maps of the C3, the C4 and the C5 stages are respectively selected to output feature maps, the feature maps of the current image or the historical image are represented, subtraction operation is carried out on the feature maps in the same stage, and the distinguishing feature information is determined based on the feature output feature maps of the C3, the C4 and the C5 stages after deletion.
Wherein, the 5 phases are respectively composed of a plurality of residual units, each residual unit is composed of a convolution layer, a normalization layer (Batch Normalization), an activation layer (relu) and an identity mapping:
The bottom layer convolution layer is responsible for extracting basic characteristics such as picture edge textures, and the high layer convolution layer is responsible for combining the texture characteristics of the bottom layer with abstraction. The normalization layer performs normalization normal distribution processing on the features. The activation layer carries out nonlinear mapping on the extracted features, and enhances the generalization capability of the model. The shortcut link is adopted in the identity mapping, so that no extra parameter is generated, the calculation complexity is not increased, and the effective return of the gradient in the subsequent twin network training is ensured.
In order to acquire the distinguishing characteristic information between the current image and the comparison image, the application introduces a twin network characteristic extraction strategy, and compared with the conventional technical means of generating a residual image by performing picture splicing or subtraction operation on two pictures based on channel dimension, the application improves the expression capability of the extracted characteristic through a twin network architecture and has strong anti-interference capability of the twin network.
In an embodiment of the present application, a possible implementation manner is provided, and determining at least one change positioning frame of a history image relative to a current image based on distinguishing feature information in step S203 may include:
(1) At least one target feature point is determined based on the distinguishing feature information.
In another possible implementation manner provided in the embodiment of the present application, the determining at least one target feature point based on the distinguishing feature information may include:
a, up-sampling distinguishing characteristic information to obtain fusion characteristic information;
and b, acquiring at least one target feature point based on the fusion feature information.
Specifically, the FPN module adopts a feature pyramid strategy to perform multi-scale and hierarchical up-sampling operation on the distinguishing feature information extracted by the backstone module, so that feature map scale amplification and feature information fusion are realized, and based on the feature output of Resnet network stages C3, C4 and C5, fusion feature output of 5 layers of the FPN modules P3, P4, P5, P6 and P7 is determined.
As shown in FIG. 5, the feature outputs of resnet network stages C3, C4 and C5 are extracted first, respectively through 1The convolution operation of step length 2 is carried out on the basis of the P5 layer, and the feature graphs of the P6 and P7 layers are respectively generated;
Then, performing top-down fusion upsampling, which can adopt nearest neighbor interpolation, adding the initial P5 level features with the initial P4 level feature map, and finally passing through 3The final P4 level characteristic output is obtained by 3 convolutions, the final P4 level characteristic is added with M3 by upsampling, and finally 3 is passed throughAnd according to the same method, the P3 level also carries out up-sampling operation through an up-sampling module, thereby realizing the effect of scale amplification and fusing necessary characteristic information through up-sampling. Wherein, the characteristic output of the P7 level corresponds to the maximum single pixel receptive field and is responsible for predicting a large area, such as the whole industrial park. The feature output of the P3 hierarchy corresponds to a single pixel receptive field 32 times smaller than P7 and is responsible for predicting small object objects such as individual buildings or bridges.
And finally, calculating regression targets corresponding to all the pixel points on all the levels, and screening out the pixel points meeting the preset conditions according to the regression targets as target characteristic points, wherein the preset conditions can be respectively set for different levels where the pixel points are located.
Through fusion up-sampling based on a feature pyramid strategy, the input of each level up-sampling module is not only from the output of the previous level up-sampling, but also from the shared feature layer with the same size corresponding to the backstone feature extraction link, and information fusion is realized by adopting convolution operation, so that better fusion of the feature information of the hierarchical level is realized. Meanwhile, the feature map output of different levels is respectively responsible for detecting target objects with different sizes, so that the accuracy of subsequent target detection is enhanced.
(2) And determining at least one change positioning frame with a preset shape based on the target characteristic points.
The embodiment of the application provides a possible implementation manner, wherein the shape of the change positioning frame can be rectangular, and the determination of the change positioning frame with at least one preset shape based on the target characteristic point can comprise the following steps:
a, respectively determining the category and margin data corresponding to each target feature point;
And b, respectively determining rectangular change positioning frames corresponding to each target feature point based on the category and the margin data.
The twin network performs classification prediction and regression on the target feature points on all levels, and determines the object category of each point and the distances between each point and four boundaries of the object, so that the category of the object and a corresponding rectangular change positioning frame can be determined.
In one possible implementation manner provided in the embodiment of the present application, updating the map based on at least one changed positioning frame in step S204 may include:
Determining a change profile based on each change positioning frame;
receiving correction information, correcting the change profile based on the correction information, and generating ground feature profile data;
And updating the map according to the ground feature profile data.
Specifically, taking building change update as an example, an accurate building contour can be generated according to a change positioning frame based on a building instance segmentation algorithm, then the building contour is edited and corrected for the second time according to manual operation, and finally the corrected building contour data is used as map update data so as to update the building on a map.
In the embodiment of the present application, a possible implementation manner is provided, as shown in fig. 6, a twin network may be obtained through training by the following steps:
S601, acquiring sample images of the same sample area in different periods, wherein the sample images are provided with corresponding sample positioning frames;
s602, inputting a sample image into an initial twin network to obtain a real-time positioning frame tag output by the initial twin network;
S603, determining a multi-branch loss function based on the sample positioning frame and the positioning frame label;
S604, adjusting parameters of the initial twin network based on the multi-branch loss function to obtain the twin network.
Specifically, the embodiment of the application uses an anchor-free strategy to directly predict targets for each pixel point, measures the distances of the upper, lower, left and right sides of a change positioning frame and the types of the targets to obtain a real-time positioning frame label, and trains a twin network based on a multi-branch loss function, wherein the multi-branch loss function comprises classification loss, regression loss and centrality loss:
(1) The classification loss can calculate the class score corresponding to the target feature point, and the class score characterizes whether the target feature point falls into a corresponding sample positioning frame or not;
(2) The regression loss can characterize the difference between the real-time positioning frame label and the sample positioning frame;
(3) Center loss in order to effectively screen the positioning frame labels, the center of each target feature point is calculated first, the center data can represent the deviation degree of the current point from the target center, and then the center data is directly multiplied by the category scores in the classification loss to be used as the final classification score.
According to the embodiment of the application, complex calculation related to the anchor is completely avoided through an anchor-free strategy, training samples of the twin network are more abundant through multi-level data processing, various folded, extremely-large, extremely-small or irregular target objects can be detected, and meanwhile, centrality loss is introduced, so that positioning frame labels generated by points far from a target center (sample positioning frame center) are effectively limited, the points with lower quality are reduced, and accuracy of twin network prediction is enhanced by using a multi-branch loss function.
In order to better understand the above map updating method, as shown in fig. 7, an example of the map updating method of the present application is described in detail below:
as shown in fig. 7, in an example, the map updating method provided by the present application may include the following steps:
S701, acquiring a current image and a historical image of a map corresponding to a region to be detected;
S702, carrying out feature extraction on a current image based on a trained twin network to obtain a first feature point set, carrying out feature extraction on a historical image to obtain a second feature point set, wherein the twin network can be an FCOS or SSD network based on a twin neural network architecture;
S703, determining a distinguishing point set between the first feature point set and the second feature point set through downsampling of the feature information, and generating distinguishing feature information based on the distinguishing point set, wherein the downsampling of the feature information can be specifically performed based on Resnet network structures;
S704, up-sampling the distinguishing characteristic information by adopting a characteristic pyramid strategy to obtain fusion characteristic information;
S705, acquiring at least one target feature point based on the fusion feature information;
s706, respectively determining the category and margin data corresponding to each target feature point;
And S707, respectively determining a rectangular change positioning frame corresponding to each target feature point based on the category and the margin data, wherein the change positioning frame characterizes a change area of the historical image relative to the current image, namely, the change condition on the same area position in the map, and the change can comprise building of a building, transformation of a road, building of a bridge and the like.
And S708, updating the map based on the at least one change positioning frame.
According to the map updating method, the distinguishing characteristic information between the current image and the historical image is determined through the twin network, the twin network is high in anti-interference capability, so that the obtained distinguishing characteristic information has better characteristic characterization capability, and the change positioning frame is determined based on the distinguishing characteristic information, so that the map is updated, the false detection of map change can be effectively reduced, and the accuracy of map updating is improved.
In addition, the change identification of the region is realized based on the change positioning frame, the map change region can be locked quickly, compared with the change identification of the pixel level of semantic segmentation in the prior art, the update efficiency of map data is greatly improved, the requirements of a network model on image resolution and image quality are reduced, and the accuracy of the change positioning of the map data is ensured.
The embodiment of the application provides a map updating device, as shown in fig. 8, the map updating device 80 may include an acquisition module 801, an extraction module 802, a determination module 803, and an updating module 804, wherein,
An obtaining module 801, configured to obtain a current image and a historical image of a map corresponding to a region to be detected;
An extraction module 802, configured to determine distinguishing feature information between the current image and the historical image based on the trained twin network;
A determining module 803 for determining at least one change localization frame of the history image with respect to the current image based on the distinguishing characteristic information;
an updating module 804 is configured to update the map based on the at least one changed positioning frame.
In one possible implementation manner provided in the embodiment of the present application, the extracting module 802 may specifically be used to:
extracting features of the current image to obtain a first feature point set;
extracting features of the historical images to obtain a second feature point set;
determining a set of distinguishing points between the first set of feature points and the second set of feature points;
based on the set of distinguishing points, distinguishing feature information is generated.
In one possible implementation manner provided in the embodiment of the present application, the determining module 803 may specifically include:
a first determining unit configured to determine at least one target feature point based on the distinguishing feature information;
and the second determining unit is used for determining at least one change positioning frame with a preset shape based on the target characteristic points.
In an embodiment of the present application, a possible implementation manner is provided, where the first determining unit may further be configured to:
upsampling the distinguishing characteristic information to obtain fusion characteristic information;
And acquiring at least one target feature point based on the fusion feature information.
In an embodiment of the present application, a possible implementation manner is provided, where the shape of the variable positioning frame may be rectangular, and the second determining unit may be further configured to:
respectively determining the category and margin data corresponding to each target feature point;
And respectively determining rectangular change positioning frames corresponding to each target feature point based on the category and the margin data.
In one possible implementation manner provided in the embodiment of the present application, the update module 804 may be further configured to:
Determining a change profile based on each change positioning frame;
receiving correction information, correcting the change profile based on the correction information, and generating ground feature profile data;
And updating the map according to the ground feature profile data.
The embodiment of the application provides a possible implementation manner, and the device further comprises a training module, which can be specifically used for:
sample images of the same sample area in different periods are acquired, and the sample images are provided with corresponding sample positioning frames;
inputting the sample image into an initial twin network to obtain a real-time positioning frame tag output by the initial twin network;
Determining a multi-branch loss function based on the sample positioning frame and the positioning frame tag;
And adjusting parameters of the initial twin network based on the multi-branch loss function to obtain the twin network.
In order to better understand the map updating apparatus described above, an example of the map updating apparatus of the present application is described in detail below, and as shown in fig. 9, an apparatus 90 of the present embodiment may include:
an acquisition module 901, configured to acquire a current image and a historical image of a map corresponding to a region to be detected;
An extracting module 902, configured to determine distinguishing feature information between the current image and the historical image based on the trained twin network;
a determining module 903 for determining at least one change positioning frame of the history image with respect to the current image based on the distinguishing characteristic information;
an update module 904 for updating the map based on the at least one changed positioning frame;
The training module 905 is configured to obtain sample images of the same sample region in different periods, set a corresponding sample positioning frame in the sample image, input the sample image to an initial twin network to obtain a real-time positioning frame tag output by the initial twin network, determine a multi-branch loss function based on the sample positioning frame and the positioning frame tag, and adjust parameters of the initial twin network based on the multi-branch loss function to obtain the twin network.
The map updating device of the present embodiment may execute the map updating method according to the above embodiment of the present application, and the implementation principle is similar, and will not be described herein.
According to the application, the distinguishing characteristic information between the current image and the historical image is determined through the twin network, so that the twin network has strong anti-interference capability, the obtained distinguishing characteristic information has better characteristic characterization capability, and the change positioning frame is determined based on the distinguishing characteristic information, so that the map is updated, the false detection of map change can be effectively reduced, and the accuracy of map updating is improved.
In addition, the change identification of the region is realized based on the change positioning frame, the map change region can be locked quickly, compared with the change identification of the pixel level of semantic segmentation in the prior art, the update efficiency of map data is greatly improved, the requirements of a network model on image resolution and image quality are reduced, and the accuracy of the change positioning of the map data is ensured.
The embodiment of the application provides electronic equipment, which comprises a memory and a processor; the method and the device have the advantages that when the method and the device are used for being executed by a processor, distinguishing characteristic information between a current image and a historical image is determined through a twin network, the twin network is high in anti-interference capability, so that the obtained distinguishing characteristic information has better characteristic characterization capability, meanwhile, region change identification is realized based on a change positioning frame, a map change area can be quickly locked, compared with change identification of pixel levels of semantic segmentation in the prior art, the updating efficiency of map data is greatly improved, the requirements of a network model on image resolution and image quality are reduced, and the accuracy of map data change positioning is guaranteed.
In an alternative embodiment, an electronic device is provided, as shown in FIG. 10, the electronic device 4000 shown in FIG. 10 comprising a processor 4001 and a memory 4003. Wherein the processor 4001 is coupled to the memory 4003, such as via a bus 4002. Optionally, the electronic device 4000 may further comprise a transceiver 4004, the transceiver 4004 may be used for data interaction between the electronic device and other electronic devices, such as transmission of data and/or reception of data, etc. It should be noted that, in practical applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 is not limited to the embodiment of the present application.
The Processor 4001 may be a CPU (Central Processing Unit ), general purpose Processor, DSP (DIGITAL SIGNAL Processor, data signal Processor), ASIC (Application SPECIFIC INTEGRATED Circuit), FPGA (Field Programmable GATE ARRAY ) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. Which may implement or perform the various exemplary logic blocks, modules and circuits described in connection with this disclosure. The processor 4001 may also be a combination that implements computing functionality, e.g., comprising one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.
Bus 4002 may include a path to transfer information between the aforementioned components. Bus 4002 may be a PCI (PERIPHERAL COMPONENT INTERCONNECT, peripheral component interconnect standard) bus or an EISA (Extended Industry Standard Architecture ) bus, or the like. The bus 4002 can be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in fig. 10, but not only one bus or one type of bus.
Memory 4003 may be, but is not limited to, ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, RAM (Random Access Memory ) or other type of dynamic storage device that can store information and instructions, EEPROM (ELECTRICALLY ERASABLE PROGRAMMABLE READ ONLY MEMORY ), CD-ROM (Compact Disc Read Only Memory, compact disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
The memory 4003 is used for storing application program codes (computer programs) for executing the present application and is controlled to be executed by the processor 4001. The processor 4001 is configured to execute application program codes stored in the memory 4003 to realize what is shown in the foregoing method embodiment.
Among them, the electronic devices include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), car terminals (e.g., car navigation terminals), and the like, and stationary terminals such as digital TVs, desktop computers, and the like. The electronic device shown in fig. 10 is merely an example and should not be construed to limit the functionality and scope of use of the disclosed embodiments.
Embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, which when run on a computer, causes the computer to perform the corresponding method embodiments described above.
It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited in order and may be performed in other orders, unless explicitly stated herein. Moreover, at least some of the steps in the flowcharts of the figures may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, the order of their execution not necessarily being sequential, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.
The foregoing is only a partial embodiment of the present invention, and it should be noted that it will be apparent to those skilled in the art that modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the present invention.