Road network road difference detection method and device, computer device and storage medium

Abstract

A method for detecting road differences in a road network, comprising: obtaining intersection image data and road mask images corresponding to the intersection image data of each road in a self-owned road network and a standard road network respectively; constructing a road matching algorithm model; training the road matching algorithm model using the intersection image data and road mask images of all roads in the self-owned road network and the standard road network; inputting intersection image data and road mask images of a road to be compared in the self-owned road network and intersection image data of the road to be compared in the standard road network; and obtaining a predicted road mask image of the road to be compared in the standard road network using the trained road matching algorithm model. The method for detecting road differences in a road network according to the embodiments of the present application uses a trained road matching algorithm model to obtain a predicted road mask image in a standard road network, has high detection efficiency and high detection accuracy, and can compare complex road differences.

Description

Technical Field

[0001] This application relates to the field of road network information technology, and in particular to a method, apparatus, computer equipment and storage medium for detecting road differences in a road network. Background Technology

[0002] With the development of road network information technology, an increasing number of network service platforms and mobile applications with navigation functions have emerged. These platforms and applications incorporate their own road network data, typically obtained from publicly available open-source road network data or different versions of existing road network data. This open-source and different versions of road network data are updated promptly based on changes in actual roads. To ensure the accuracy of the proprietary road network data, when these data are updated, it is necessary to perform difference detection between the proprietary and open-source road networks, thereby updating and correcting the proprietary road network data accordingly.

[0003] Currently, the common method for detecting road network differences is differential detection. This involves performing differential analysis on road data from open-source road network data and different versions of road network data, and then confirming the differences to correct the road data and improve the quality of proprietary road network data.

[0004] In differential calculations, geometric information, attribute information, and relationship information with adjacent roads of the road segment to be compared are typically used to set corresponding differential rules for manual matching, resulting in the detection results of road differences. This method is relatively simple to implement and can only detect relatively simple road differences, such as the addition or deletion of roads. When the two versions of road data have more complex differences, such as overall or partial positional shifts in roads, or significant differences in road morphology, differential methods cannot accurately compare the differences between roads.

[0005] In view of the problems existing in the existing methods, there is an urgent need to provide a method, device, computer equipment and storage medium for detecting road differences in road networks that can compare and identify complex road differences. Summary of the Invention

[0006] The purpose of this application is to provide a method, apparatus, computer equipment, and storage medium for detecting road network differences that can compare complex road differences, in order to solve one of the aforementioned technical problems.

[0007] In a first aspect, a method for detecting road differences in a road network includes: acquiring intersection image data of each road in a proprietary road network and a standard road network, and road mask maps corresponding to the intersection image data; constructing a road matching algorithm model; training the road matching algorithm model using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks; inputting the intersection image data and road mask maps of the road to be compared in the proprietary road network, and the intersection image data of the road to be compared in the standard road network, and using the trained road matching algorithm model to obtain a predicted road mask map of the road to be compared in the standard road network.

[0008] Preferably, after the step of obtaining the predicted road mask map of the road to be compared in the standard road network using the trained road matching algorithm model, the method further includes: obtaining the prediction difference detection result between the predicted road mask map and the road mask map in the proprietary road network using the trained road matching algorithm model.

[0009] Preferably, the steps of obtaining intersection image data and corresponding road mask maps for each road in the proprietary road network and the standard road network are as follows: obtaining road trajectory data of the roads in the proprietary road network; dividing the road trajectory data into small segments of road segments to be compared, based on intersections; obtaining road mask maps corresponding to the road segments to be compared at each intersection in the proprietary road network; expanding the road segments to be compared outwards by a certain range in the proprietary road network to obtain intersection image data; and obtaining the relative intersection image data of the road segments to be compared on the standard road network. The standard road segment is determined; it is then determined whether a corresponding standard road segment exists in the standard road network: if not, the similar road category is marked as 0; if yes, it is further determined whether the road shape of the standard road segment is similar to that of the road segment to be compared in the proprietary road network; if similar, the similar road category is marked as 1, and the intersection image data and road mask map of the standard road segment are obtained in the standard road network; if not similar, the similar road category is marked as 0, and the intersection image data and road mask map of the standard road segment are obtained in the standard road network.

[0010] Preferably, the steps for constructing the road matching algorithm model described above are as follows: inputting the intersection image data and the road mask image of the proprietary road network into a first encoding network module for feature extraction to obtain a first feature vector; inputting the first feature vector into two convolutional computation layers to obtain a first matrix and a second matrix; inputting the intersection image data of the standard road network into a second encoding network module for feature extraction to obtain a second feature vector; upsampling and downsampling the road mask image of the proprietary road network to the size of the second feature vector to obtain a sampled road mask image; and then inputting the second feature vector and the sampled road mask image into a second encoding network module for feature extraction to obtain a second feature vector. After the road mask images are stitched together, they are input into the two convolutional computation layers to obtain a third matrix and a fourth matrix. The first matrix, the second matrix, and the third matrix are multiplied together and then stitched together with the fourth matrix to obtain a stitched matrix. The stitched matrix is ​​input into the decoding network module to obtain a predicted road mask image within the standard road network. The difference between the predicted road mask image within the standard road network and the road mask image within the standard road network is compared. Based on the difference gradient backpropagation, the convolution kernel parameters of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, are optimized respectively.

[0011] Preferably, after the steps of optimizing the convolution kernel parameters of the two convolutional computation layers according to the difference gradient backpropagation, the method includes: a step of comparing the similarity of the predicted road mask; the step of comparing the similarity of the predicted road mask specifically includes: setting a similarity threshold; concatenating the predicted road mask and the intersection image data of the standard road network and inputting them into the first encoding network module for feature extraction to obtain a third feature vector; calculating the Euclidean distance between the first feature vector and the third feature vector; determining whether the Euclidean distance is greater than the similarity threshold: if no, the prediction difference detection result is 1, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is small; if yes, the prediction difference detection result is 0, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is large; comparing whether the prediction difference detection result and the similar road category are consistent, if they are inconsistent, optimizing the parameters of the road matching algorithm model.

[0012] Preferably, after the step where the predicted difference detection result is 0, indicating a significant difference between the predicted road mask and the road mask in the proprietary road network, the method further includes: a step of further verifying the difference between the predicted road mask and the road mask in the proprietary road network; the step of further verifying the difference between the predicted road mask and the road mask in the proprietary road network specifically includes: obtaining a first latitude and longitude point sequence of the road segment to be compared in the proprietary road network; obtaining a second latitude and longitude point sequence of the standard road segment in the predicted road mask of the standard road network; and comparing the difference between the first latitude and longitude point sequence and the second latitude and longitude point sequence.

[0013] Preferably, training the road matching algorithm model using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks specifically involves: performing road matching training and similarity comparison training on the road matching algorithm model using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks; specifically, performing road matching training and similarity comparison training on the road matching algorithm model using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks involves: performing road matching training on the road matching algorithm model using the binary classification cross-entropy loss function method and similarity comparison training using the contrastive loss function method; the first encoding network module, the second encoding network module, and the decoding network module can be residual networks or visual geometric group networks.

[0014] Secondly, embodiments of this application provide a road network road difference detection device, comprising: a data construction module for acquiring intersection image data of each road in a proprietary road network and a standard road network, and road mask maps corresponding to the intersection image data; an algorithm construction module for constructing a road matching algorithm model; a model training module for training the road matching algorithm model using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks; and a result calculation module for inputting the intersection image data and road mask maps of the road to be compared in the proprietary road network, and the intersection image data of the road to be compared in the standard road network, and obtaining a predicted road mask map of the road to be compared in the standard road network using the trained road matching algorithm model.

[0015] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for detecting road network differences as described above.

[0016] Fourthly, embodiments of this application provide a computer device, including: one or more processors; a memory; and one or more computer programs, wherein the processors and the memory are connected via a bus, and wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, wherein when the processors execute the computer programs, they implement the steps of the method for detecting road network differences as described above.

[0017] The road network road difference detection method in this application first acquires the intersection image data and corresponding road mask maps of each road in its own road network and the standard road network. That is, the intersection image data and road mask map of a road in its own road network, and the intersection image data and road mask map of a road in the standard road network. Figure 4 Each map serves as a set of training data, and multiple roads have multiple sets of training data. To ensure the accuracy of subsequent training of the road matching algorithm model, the training data can be a large amount of data containing various scenarios.

[0018] Owned road networks can be network service platforms with navigation functions or mobile applications that require improvement. Standard road networks are existing technologies that can be used for comparison and reference by owned road networks. Owned road networks can update themselves by referring to and comparing with standard road networks. Standard road networks can be road networks of the same or different origin as owned road networks, or they can be road networks of the same origin but different versions. Specifically, standard road networks can be open-source road networks, which can be existing public maps such as OpenStreetMap (OSM). This public map is an online map collaboration project with the goal of creating a world map with free content that can be edited by everyone.

[0019] The road network road difference detection method of this application embodiment constructs a road matching algorithm model and trains it using intersection image data and road mask maps of all proprietary road networks and standard road networks. Finally, it inputs the intersection image data and road mask maps of the road to be compared in the proprietary road network and the intersection image data of the road to be compared in the standard road network, and uses the trained road matching algorithm model to obtain a predicted road mask map of the road to be compared within the standard road network. In other words, compared with existing technologies that only perform simple difference comparisons of road data, the road network road difference detection method of this application embodiment uses a trained road matching algorithm model to obtain a predicted road mask map within the standard road network, resulting in high detection efficiency, high detection accuracy, and the ability to identify complex road differences. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for detecting road network differences according to an embodiment of this application;

[0021] Figure 2 This is a flowchart of a method for obtaining intersection image data and corresponding road mask maps of each road in a proprietary road network and a standard road network for road network difference detection, provided in an embodiment of this application.

[0022] Figure 3 This is a flowchart of a method for constructing a road matching algorithm model for road network road difference detection provided in an embodiment of this application;

[0023] Figure 4 This is a flowchart of a method for comparing the similarity of predicted road mask images in road network difference detection according to an embodiment of this application;

[0024] Figure 5 This is a flowchart of a method for further verifying and predicting the difference between a road mask map and a road mask map in an existing road network, provided in an embodiment of this application;

[0025] Figure 6 This is a schematic diagram of the structure of a road network road difference detection device provided in an embodiment of this application;

[0026] Figure 7 This is a structural block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] To illustrate the technical solution of this application, specific embodiments are described below.

[0029] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0030] To illustrate the technical solution of this application, specific embodiments are described below.

[0031] Please see Figure 1 This is a flowchart of a method for detecting road network differences according to an embodiment of this application. This application provides a method for detecting road network differences, including:

[0032] S101. Obtain the intersection image data of each road in the proprietary road network and the standard road network, and the road mask map corresponding to the intersection image data;

[0033] S102. Construct a road matching algorithm model;

[0034] S103. Train the road matching algorithm model using intersection image data and road mask maps of all proprietary road networks and standard road networks;

[0035] S104. Input the intersection image data and road mask map of the road to be compared in the proprietary road network, and the intersection image data of the road to be compared in the standard road network. Use the trained road matching algorithm model to obtain the predicted road mask map of the road to be compared in the standard road network.

[0036] A mask is a two-dimensional matrix represented by 0-1 binary values. A road mask has values ​​of 1 for the road trajectory and 0 for other areas. A road mask can represent the trajectory of a road. Image data refers to the collection of grayscale values ​​of each pixel, used to represent an image.

[0037] The road network road difference detection method in this application first acquires the intersection image data and corresponding road mask maps of each road in its own road network and the standard road network. That is, the intersection image data and road mask map of a road in its own road network, and the intersection image data and road mask map of a road in the standard road network. Figure 4 Each map serves as a set of training data, and multiple roads have multiple sets of training data. To ensure the accuracy of subsequent training of the road matching algorithm model, the training data can be a large amount of data containing various scenarios.

[0038] Owned road networks can be network service platforms with navigation functions or mobile applications that require improvement. Standard road networks are existing technologies that can be used for comparison and reference by owned road networks. Owned road networks can update themselves by referring to and comparing with standard road networks. Standard road networks can be road networks of the same or different origin as owned road networks, or they can be road networks of the same origin but different versions. Specifically, standard road networks can be open-source road networks, which can be existing public maps such as OpenStreetMap (OSM). This public map is an online map collaboration project with the goal of creating a world map with free content that can be edited by everyone.

[0039] The road network road difference detection method of this application embodiment constructs a road matching algorithm model and trains it using intersection image data and road mask maps of all proprietary road networks and standard road networks. Specifically, the road matching algorithm model undergoes deep learning training with a large amount of training data. Finally, it inputs the intersection image data and road mask maps of the road to be compared in the proprietary road network and the intersection image data of the road to be compared in the standard road network. The trained road matching algorithm model then obtains a predicted road mask map of the road to be compared within the standard road network. In other words, compared to existing technologies that only perform simple difference comparisons of road data, the road network road difference detection method of this application embodiment uses a trained road matching algorithm model to obtain a predicted road mask map within the standard road network, avoiding a large amount of manual comparison and difference detection work. This results in high detection efficiency, high detection accuracy, and the ability to identify complex road differences.

[0040] Preferably, after obtaining the predicted road mask map of the road to be compared within the standard road network using the trained road matching algorithm model, the process further includes:

[0041] The trained road matching algorithm model is used to obtain the prediction difference detection results between the predicted road mask map and the predicted road mask map in the proprietary road network.

[0042] The road network road difference detection method of this application embodiment can obtain the prediction difference detection results of the predicted road mask map and the road mask map in the self-owned road network through the trained road matching algorithm model. Specifically, the value of the prediction difference detection result can be set to 0 and 1. When the prediction difference detection result is 0, it indicates that the difference between the predicted road mask map and the road mask map in the self-owned road network is large or no corresponding prediction road mask map is found in the standard road network. When the prediction difference detection result is 1, it indicates that the difference between the predicted road mask map and the road mask map in the self-owned road network is small, that is, the road shape difference of the road segment to be detected is small in the self-owned road network and the standard road network, and the road shape is similar.

[0043] The embodiments of this application use a trained road matching algorithm model to obtain the prediction difference detection results between the predicted road mask map and the predicted road mask map in the self-owned road network, which does not require manual comparison of differences and is highly efficient.

[0044] Please see Figure 2 This is a flowchart illustrating a method for obtaining intersection image data and corresponding road mask maps of each road in both a proprietary road network and a standard road network, according to an embodiment of this application. The method includes the following steps:

[0045] S201. Obtain road trajectory data of the self-owned road network;

[0046] S202. Divide the road trajectory data into smaller segments of road sections to be compared, based on intersections.

[0047] S203. Obtain the road mask map corresponding to the road segment to be compared at each intersection in the proprietary road network;

[0048] S204. Expand the road segment to be compared to a certain range on the existing road network to obtain intersection image data;

[0049] S205. Obtain the standard road segments corresponding to the road segments to be compared from the standard road network;

[0050] S206. Determine whether a corresponding standard road segment exists in the standard road network; if not, proceed to step S210; if yes, proceed to step S207.

[0051] S207. Further determine whether the road shape of the standard road segment and the road segment to be compared in the proprietary road network are similar: if they are similar, proceed to step S208; if they are not similar, proceed to step S209.

[0052] S208, label similar roads as category 1, and obtain intersection image data and road mask maps of standard road segments from the standard road network;

[0053] S209. Mark similar road categories as 0, and obtain intersection image data and road mask maps of standard road segments from the standard road network;

[0054] S210, similar road category is marked as 0.

[0055] Road trajectory data can be an ordered string of latitude and longitude coordinates representing road trajectories within a road network. The road trajectory data is then divided into smaller segments for comparison, organized by intersection. For example, a relatively long road is broken down into multiple smaller segments for comparison. Within the existing road network, the comparison road segments are expanded outwards by a certain range to obtain intersection image data. Specific expansion rules can be set as needed, such as expanding in a square, circular, or irregular shape.

[0056] This application embodiment obtains intersection image data and corresponding road mask maps for each road in both the proprietary road network and the standard road network through the above steps. That is, it obtains data for the road segments to be compared on a per-intersection basis. Each road segment to be compared includes a road mask map and intersection image data from the proprietary road network, as well as a road mask map and intersection image data from the standard road network. These four sets of image data serve as a set of training data. Multiple road segments to be compared have multiple sets of training data, used for subsequent road matching algorithm model training. Similar road categories are labeled to provide ground truth for subsequent similarity comparison training of the road matching algorithm model.

[0057] Please see Figure 3 This is a flowchart of a method for constructing a road matching algorithm model for road network road difference detection according to one embodiment of this application. Another embodiment of this application provides a method for constructing a road matching algorithm model for road network road difference detection, which includes the following steps:

[0058] S301. Input the intersection image data and road mask map of the self-owned road network into the first coding network module to extract features and obtain the first feature vector;

[0059] S302. Input the first feature vector into two convolutional computation layers respectively to obtain the first matrix and the second matrix;

[0060] S303. Input the intersection image data of the standard road network into the second coding network module to extract features and obtain the second feature vector;

[0061] S304. After upsampling and downsampling the road mask image of the self-owned road network to the size of the second feature vector, the sampled road mask image is obtained.

[0062] S305. The second feature vector and the sampled road mask are concatenated and then input into two convolutional computation layers to obtain the third matrix and the fourth matrix.

[0063] S306. Multiply the first matrix, the second matrix, and the third matrix together and then concatenate them with the fourth matrix to obtain the concatenated matrix;

[0064] S307. Input the splicing matrix into the decoding network module to obtain the predicted road mask map in the standard road network;

[0065] S308. Compare the differences between the predicted road mask map and the road mask map within the standard road network;

[0066] S309. Optimize the convolution kernel parameters of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, respectively, based on the differential gradient backpropagation.

[0067] Specifically, the first encoding network module, the second encoding network module, and the decoding network module can be any commonly used network in the existing image processing field, such as ResNet or VGG. These modules can use networks with the same structure, but their network parameters are not shared and they need to be trained separately. ResNet is a residual network based on existing technology, constructed from residual building blocks. VGG (Visual Geometry Group Network) is a visual geometry group network based on existing technology.

[0068] Convolutional layers are used for convolution calculations and are a fundamental component of existing deep learning training models. They are special layers where convolutional kernels can be applied. Convolution is one of the most widely used techniques in deep learning, used to extract features from images, thereby enhancing the accuracy of the learning model. The convolution kernel is the parameter used in the convolution operation, the core of the computational operation that transforms pixel values ​​in an image into features. The convolutional layer takes the content of the training image from the input layer as input and the convolution kernel as a parameter. During computation, the convolution kernel is multiplied with pixel attributes such as edges, contours, and colors of the image, thereby extracting image features and improving the accuracy of the deep learning model.

[0069] This application embodiment compares the differences between the predicted road mask map and the road mask map within the standard road network, and optimizes the convolution kernel parameters of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, based on the gradient backpropagation of the differences. This allows for continuous training and deep learning of the road matching algorithm model, continuously optimizing the convolution kernel parameters of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module. This continuous optimization of the road matching algorithm model ensures the accuracy of the predicted road mask map within the standard road network.

[0070] The embodiments of this application construct a road matching algorithm model using the above method, providing an algorithm model foundation for subsequent model training and deep learning.

[0071] Preferably, after the step of optimizing the convolution kernel parameters of the two convolutional computation layers based on the differential gradient backpropagation, the method includes a step of comparing the similarity of the predicted road mask images. That is, based on obtaining the predicted road mask images within the standard road network, a further similarity comparison is performed on the predicted road mask images, and the road data of the proprietary road network can be improved based on the similarity comparison results.

[0072] Please see Figure 4This is a flowchart of a method for comparing the similarity of predicted road mask images in road network difference detection according to an embodiment of this application. The method for comparing the similarity of predicted road mask images in road network difference detection according to an embodiment of this application includes the following steps:

[0073] S401. Set the similarity threshold;

[0074] S402. The predicted road mask map and the intersection image data of the standard road network are stitched together and then input into the first coding network module for feature extraction to obtain the third feature vector;

[0075] S403. Calculate the Euclidean distance between the first eigenvector and the third eigenvector;

[0076] S404. Determine whether the Euclidean distance is greater than the similarity threshold: If no, the prediction difference detection result is 1, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is small; if yes, the prediction difference detection result is 0, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is large.

[0077] S405. Compare the predicted difference detection results with the similar road categories to see if they are consistent. If they are inconsistent, optimize the parameters of the road matching algorithm model.

[0078] Euclidean distance is a commonly used definition of distance, referring to the true distance between two points in m-dimensional space, or the natural length of a vector (i.e., the distance from that point to the origin). In two-dimensional and three-dimensional space, Euclidean distance is simply the actual distance between two points.

[0079] The parameters of the road matching algorithm model may include the convolution kernel parameters of the two convolutional computation layers, the similarity threshold, and the parameters of the first encoding network module, the second encoding network module, and the decoding network module.

[0080] The similarity threshold can be set according to actual needs.

[0081] The embodiments of this application optimize the parameters of the road matching algorithm model through the above steps, providing a foundation for subsequent comparison of the similarity of predicted road mask images.

[0082] Preferably, after the step where the prediction difference detection result is 0, indicating a significant difference between the predicted road mask and the road mask in the proprietary road network, the method further includes: a step of further verifying the difference between the predicted road mask and the road mask in the proprietary road network. That is, when the road matching algorithm model predicts a significant difference between the predicted road mask and the road mask in the proprietary road network, further verification of this difference can effectively avoid errors in the road matching algorithm model, thereby further improving the accuracy of the predicted road mask.

[0083] Please see Figure 5 This is a flowchart illustrating a method for further verifying the difference between a predicted road mask and a road mask in an existing road network, according to an embodiment of this application. The method includes the following steps:

[0084] S501. Obtain the first latitude and longitude point sequence of the road segment to be compared in the self-owned road network;

[0085] S502. Obtain the second latitude and longitude point sequence of the standard road segments in the predicted road mask map of the standard road network;

[0086] S503. Compare the differences between the first latitude and longitude point sequence and the second latitude and longitude point sequence.

[0087] Road network data is typically represented by ordered latitude and longitude point sequences. By comparing the differences between the first and second latitude and longitude point sequence sequences, the difference between the predicted road mask and the road mask in the existing road network can be further determined. This effectively avoids errors in the road matching algorithm model and thus improves the accuracy of the predicted road mask. Specifically, the differences between the predicted road mask and the road mask in the existing road network can be further verified manually or by computer programs.

[0088] Preferably, training the road matching algorithm model using intersection image data and road mask maps of all proprietary road networks and standard road networks specifically involves: using intersection image data and road mask maps of all proprietary road networks and standard road networks to perform road matching training and similarity comparison training on the road matching algorithm model.

[0089] Specifically, road matching training can optimize the convolution kernel parameters of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, respectively, based on the difference between the predicted road mask map in the standard road network and the road mask map in the standard road network.

[0090] Similarity comparison training can compare whether the predicted difference detection results are consistent with the similar road categories. If they are inconsistent, the parameters of the road matching algorithm model can be optimized.

[0091] Further preferably, the road matching algorithm model is trained for road matching and similarity comparison using intersection image data and road masks from all proprietary and standard road networks. Specifically, the road matching algorithm model is trained for road matching using the binary cross-entropy loss function and for similarity comparison using the contrastive loss function. Specifically, the binary cross-entropy loss function (BCELoss) and the contrastive loss function (ContrastiveLoss) are existing techniques.

[0092] Please see Figure 6 A road network road difference detection device 10, comprising:

[0093] Data construction module 11 is used to obtain the intersection image data of each road in its own road network and the standard road network, and the road mask map corresponding to the intersection image data.

[0094] Algorithm building module 12 is used to build the road matching algorithm model;

[0095] Model training module 13 is used to train a road matching algorithm model using intersection image data and road mask maps of all proprietary road networks and standard road networks;

[0096] The result calculation module 14 is used to input the intersection image data and road mask map of the road to be compared in the self-owned road network, as well as the intersection image data of the road to be compared in the standard road network, and to obtain the predicted road mask map of the road to be compared in the standard road network using the trained road matching algorithm model.

[0097] The road network road difference detection device 10 provided in one embodiment of this application and the road network road difference detection method provided in the above embodiment of this application belong to the same concept. The specific implementation process is detailed in the full text of the specification and will not be repeated here.

[0098] An embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the road network difference detection method provided in the above embodiments of this application.

[0099] Figure 7 This document illustrates a specific structural block diagram of a computer device provided in an embodiment of this application. The computer device 100 includes one or more processors 101, a memory 102, and one or more computer programs. The processors 101 and the memory 102 are connected via a bus. The one or more computer programs are stored in the memory 102 and configured to be executed by the one or more processors 101. When the processors 101 execute the computer programs, they implement the steps of the method for detecting road network differences as provided in the embodiments of this application described above.

[0100] Computer devices can be desktop computers, mobile terminals, etc. Mobile terminals include mobile phones, tablets, laptops, personal digital assistants, etc.

[0101] The road network road difference detection method in this application first acquires the intersection image data and corresponding road mask maps of each road in its own road network and the standard road network. That is, the intersection image data and road mask map of a road in its own road network, and the intersection image data and road mask map of a road in the standard road network. Figure 4 Each map serves as a set of training data, and multiple roads have multiple sets of training data. To ensure the accuracy of subsequent training of the road matching algorithm model, the training data can be a large amount of data containing various scenarios.

[0102] Owned road networks can be network service platforms with navigation functions or mobile applications that require improvement. Standard road networks are existing technologies that can be used for comparison and reference by owned road networks. Owned road networks can update themselves by referring to and comparing with standard road networks. Standard road networks can be road networks of the same or different origin as owned road networks, or they can be road networks of the same origin but different versions. Specifically, standard road networks can be open-source road networks, which can be existing public maps such as OpenStreetMap (OSM). This public map is an online map collaboration project with the goal of creating a world map with free content that can be edited by everyone.

[0103] The road network road difference detection method of this application embodiment constructs a road matching algorithm model and trains it using intersection image data and road mask maps of all proprietary road networks and standard road networks. Finally, it inputs the intersection image data and road mask maps of the road to be compared in the proprietary road network and the intersection image data of the road to be compared in the standard road network, and uses the trained road matching algorithm model to obtain a predicted road mask map of the road to be compared within the standard road network. In other words, compared with existing technologies that only perform simple difference comparisons of road data, the road network road difference detection method of this application embodiment uses a trained road matching algorithm model to obtain a predicted road mask map within the standard road network, resulting in high detection efficiency, high detection accuracy, and the ability to identify complex road differences.

[0104] It should be understood that the steps in the various embodiments of this application are not necessarily executed sequentially according to the order indicated by the step numbers. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0105] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be accomplished by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), and enhanced SDRAM.

[0106] ESDRAM, Synchlink DRAM (SLDRAM), Rambus Direct RAM (RDRAM), Direct Memory Bus Dynamic RAM (DRDRAM), and Memory Bus Dynamic RAM (RDRAM), etc.

[0107] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0108] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for detecting road network differences, characterized in that, include: Obtain the intersection image data of each road in the proprietary road network and the standard road network, and the road mask map corresponding to the intersection image data; Construct a road matching algorithm model; The road matching algorithm model is trained using the intersection image data of all the proprietary road networks and the standard road networks and the road mask map; Input the intersection image data and road mask map of the road to be compared in the proprietary road network, and the intersection image data of the road to be compared in the standard road network. Use the trained road matching algorithm model to obtain the predicted road mask map of the road to be compared in the standard road network. The specific steps for constructing the road matching algorithm model are as follows: The intersection image data and the road mask map of the proprietary road network are input into the first coding network module for feature extraction to obtain the first feature vector; The first feature vector is input into two convolutional layers to obtain the first matrix and the second matrix; The intersection image data of the standard road network is input into the second coding network module for feature extraction to obtain the second feature vector; The road mask image of the proprietary road network is upsampled or downsampled to the size of the second feature vector to obtain the sampled road mask image; The second feature vector and the sampled road mask are concatenated and then input into the two convolutional computation layers to obtain the third matrix and the fourth matrix, respectively. Multiply the first matrix, the second matrix, and the third matrix together and then concatenate them with the fourth matrix to obtain a concatenated matrix; The stitching matrix is ​​input into the decoding network module to obtain a predicted road mask map within the standard road network; Compare the predicted road mask map within the standard road network with the road mask map within the standard road network to identify the differences. Based on the differential gradient backpropagation, the parameters of the convolution kernels of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, are optimized respectively.

2. The method for detecting road network differences as described in claim 1, characterized in that, After the step of obtaining the predicted road mask map of the road to be compared within the standard road network using the trained road matching algorithm model, the method further includes: The predicted difference detection results between the predicted road mask map and the road mask map in the proprietary road network are obtained using the trained road matching algorithm model.

3. The method for detecting road network differences as described in claim 2, characterized in that, The specific steps for obtaining the intersection image data of each road in the proprietary road network and the standard road network, and the road mask map corresponding to the intersection image data, are as follows: Obtain road trajectory data of the proprietary road network; The road trajectory data is divided into small segments of road sections to be compared, with each intersection as a unit. Obtain road mask images corresponding to the road segments to be compared at each intersection in the proprietary road network; The road segment to be compared is expanded outward by a certain range from the existing road network to obtain intersection image data. Obtain the standard road segment corresponding to the road segment to be compared on the standard road network; Determine whether a corresponding standard road segment exists in the standard road network: if not, mark the similar road category as 0; if yes, further determine whether the road shape of the standard road segment is similar to that of the road segment to be compared in the proprietary road network; if similar, mark the similar road category as 1, and obtain the intersection image data and road mask map of the standard road segment in the standard road network; if not similar, mark the similar road category as 0, and obtain the intersection image data and road mask map of the standard road segment in the standard road network.

4. The method for detecting road network differences as described in claim 3, characterized in that, The step of optimizing the convolution kernel parameters of the two convolutional computation layers according to the differential gradient backpropagation includes: a step of comparing the similarity of the predicted road mask map; The specific steps for performing similarity comparison on the predicted road mask image are as follows: Set a similarity threshold; The predicted road mask and the intersection image data of the standard road network are stitched together and then input into the first coding network module for feature extraction to obtain the third feature vector; Calculate the Euclidean distance between the first feature vector and the third feature vector; Determine whether the Euclidean distance is greater than the similarity threshold: if no, the predicted difference detection result is 1, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is small; if yes, the predicted difference detection result is 0, indicating that the difference between the predicted road mask and the road mask in the proprietary road network is large. Compare the predicted difference detection results with the similar road categories to see if they are consistent. If they are inconsistent, optimize the parameters of the road matching algorithm model.

5. The method for detecting road network differences as described in claim 4, characterized in that, The step following the step where the prediction difference detection result is 0, indicating a significant difference between the predicted road mask map and the road mask map in the proprietary road network, further includes: The step of further verifying the difference between the predicted road mask map and the road mask map in the proprietary road network; The step of further verifying the difference between the predicted road mask map and the road mask map in the proprietary road network specifically includes: Obtain the first latitude and longitude point sequence of the road segment to be compared in the proprietary road network; Obtain the second latitude and longitude point sequence of the standard road segment in the predicted road mask map of the standard road network; Compare the differences between the first latitude and longitude point sequence and the second latitude and longitude point sequence.

6. The method for detecting road network differences as described in claim 5, characterized in that, The step of training the road matching algorithm model using the intersection image data and road mask map of all the proprietary road networks and the standard road networks specifically involves: performing road matching training and similarity comparison training on the road matching algorithm model using the intersection image data and road mask map of all the proprietary road networks and the standard road networks; The road matching algorithm model is trained using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks. Specifically, the road matching algorithm model is trained using the intersection image data and road mask maps of all the proprietary road networks and the standard road networks, and trained using the contrastive loss function method for similarity comparison. The first encoding network module, the second encoding network module, and the decoding network module are residual networks or visual geometric group networks.

7. A road network road difference detection device, comprising: The data construction module is used to acquire the intersection image data of each road in the proprietary road network and the standard road network, and the road mask map corresponding to the intersection image data; The algorithm building module is used to construct road matching algorithm models; The model training module is used to train the road matching algorithm model using the intersection image data of all the proprietary road networks and the standard road networks and the road mask map; The result calculation module is used to input the intersection image data and road mask map of the road to be compared in the proprietary road network, as well as the intersection image data of the road to be compared in the standard road network, and to obtain the predicted road mask map of the road to be compared in the standard road network using the trained road matching algorithm model. The specific steps for constructing the road matching algorithm model are as follows: The intersection image data and the road mask map of the proprietary road network are input into the first coding network module for feature extraction to obtain the first feature vector; The first feature vector is input into two convolutional layers to obtain the first matrix and the second matrix; The intersection image data of the standard road network is input into the second coding network module for feature extraction to obtain the second feature vector; The road mask image of the proprietary road network is upsampled or downsampled to the size of the second feature vector to obtain the sampled road mask image; The second feature vector and the sampled road mask are concatenated and then input into the two convolutional computation layers to obtain the third matrix and the fourth matrix, respectively. Multiply the first matrix, the second matrix, and the third matrix together and then concatenate them with the fourth matrix to obtain a concatenated matrix; The stitching matrix is ​​input into the decoding network module to obtain a predicted road mask map within the standard road network; Compare the predicted road mask map within the standard road network with the road mask map within the standard road network to identify the differences. Based on the differential gradient backpropagation, the parameters of the convolution kernels of the two convolutional computation layers, as well as the parameters of the first encoding network module, the second encoding network module, and the decoding network module, are optimized respectively.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for detecting road network differences as described in any one of claims 1 to 6.

9. A computer device, comprising: One or more processors; Memory; as well as One or more computer programs, wherein the processor and the memory are connected via a bus, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, characterized in that, when the processor executes the computer programs, it implements the steps of the method for detecting road network differences as described in any one of claims 1 to 6.

Images