Method, device, electronic device and storage medium for determining traffic topology map

Abstract

The present application provides a method, device, electronic device and storage medium for determining a traffic topology map, the method comprising: obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment, performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by using the navigation map as prior information and fine-tuning the lane recognition model according to the image; performing traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result; performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map. The present application recognizes images by using a navigation map with lane information as a prior information target lane recognition model, thereby improving the accuracy of lane recognition results, enhancing the topological reasoning ability of the vehicle in complex traffic scenarios, and improving the accuracy and robustness of the traffic topology map.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and more specifically, to a method, device, electronic device and storage medium for determining a traffic topology map. Background Art

[0002] With the development of science and technology and the popularization of automobiles, the intelligent driving system of automobiles needs to cope with dynamic and complex driving environments to control the driving of vehicles. The complex driving environment can be diverse scenes including urban streets, highways, intersections, etc. In these scenes, the car must not only identify roads and lanes, but also understand the topological relationship between these elements. For example, at an intersection, the vehicle must understand the connection relationship between lanes to make correct path planning and decisions.

[0003] At present, in order to generate traffic topology maps for automobile intelligent driving systems, they rely heavily on high-precision maps, which can provide detailed lanes, traffic signs and other key elements with precise location information. However, the construction and maintenance process of high-precision maps not only requires a lot of human resources, but also is extremely costly. The updating and expansion of such maps also face huge challenges, especially in rapidly changing urban environments. This high cost and complexity severely limits the widespread application and expansion capabilities of autonomous driving technology. Therefore, how to generate traffic topology maps without relying on high-precision maps has become an urgent problem to be solved. Summary of the invention

[0004] In view of this, the embodiments of the present application propose a method, device, electronic device and storage medium for determining a traffic topology map to improve the above-mentioned problems.

[0005] According to a first aspect of an embodiment of the present application, a method for determining a traffic topology map is provided, the method comprising: acquiring an image of an environment in which a vehicle is located and a navigation map corresponding to the environment; performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by using the navigation map as prior information and fine-tuning the lane recognition model according to the image; performing traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result; performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0006] According to a second aspect of an embodiment of the present application, a device for determining a traffic topology map is provided, the device comprising: a navigation map acquisition module, for acquiring an image of an environment in which a vehicle is located and a navigation map corresponding to the environment; a lane recognition module, for performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by using the navigation map as prior information and fine-tuning the lane recognition model according to the image; a traffic element recognition module, for performing traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result; and a topology analysis module, for performing topology analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0007] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method for determining a traffic topology map as described above is implemented.

[0008] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the method for determining a traffic topology map as described above is implemented.

[0009] In the scheme of the present application, the target lane recognition model is obtained by using the navigation map corresponding to the environment where the vehicle is located as prior information, and the lane recognition model is fine-tuned by using the image of the environment where the vehicle is located to obtain the target lane recognition model, and then the lane recognition is performed on the image of the environment where the vehicle is located by the target lane recognition model with prior information to obtain the lane recognition result, and the traffic element recognition is performed on the image by the traffic element recognition model to obtain the traffic element recognition result, and finally a topological analysis is performed based on the lane recognition result to obtain a traffic topology map. The present application obtains accurate lane recognition results by using the navigation map with lane information as prior information to recognize the image of the environment where the vehicle is located by using the target lane recognition model, and the traffic element recognition result is obtained by recognizing the image of the environment where the vehicle is located by using the traffic element recognition model, so that a traffic topology map can be obtained by performing a topological analysis based on the accurate lane recognition results and the traffic element recognition results, thereby enhancing the topological reasoning ability of the vehicle in complex traffic scenes and improving the accuracy and robustness of the traffic topology map.

[0010] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 It is a flow chart of a method for determining a traffic topology map according to an embodiment of the present application.

[0013] Figure 2 It is a flow chart of a method for determining a traffic topology map according to another embodiment of the present application.

[0014] Figure 3 It is a schematic diagram of training a lane recognition model according to an embodiment of the present application.

[0015] Figure 4 It is a schematic diagram of the specific step flow of step 240 according to an embodiment of the present application.

[0016] Figure 5 It is a flowchart of a method for determining a traffic topology map according to another embodiment of the present application.

[0017] Figure 6 It is a schematic diagram of the specific step flow of step 370 according to an embodiment of the present application.

[0018] Figure 7 It is a flow chart of a method for determining a traffic topology map according to another embodiment of the present application.

[0019] Figure 8 It is a flow chart of a method for determining a traffic topology map according to yet another embodiment of the present application.

[0020] Fig. 9 It is a flow chart of a method for determining a traffic topology map according to another embodiment of the present application.

[0021] Fig.10 It is a block diagram of a device for determining a traffic topology map according to an embodiment of the present application.

[0022] Fig.11 It is a hardware structure diagram of an electronic device according to an embodiment of the present application.

[0023] The above-mentioned drawings have shown clear embodiments of the present invention, which will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention to computer technicians in this field through specific embodiments. DETAILED DESCRIPTION

[0024] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] See also Figure 1 , Figure 1 The method for determining a traffic topology map provided by an embodiment of the present application is shown. In a specific embodiment, the method for determining a traffic topology map can be applied to Fig.10 The traffic topology map determining device 600 and the electronic device 700 equipped with the traffic topology map determining device 600 ( Fig.11 The specific process of this embodiment will be described below. Of course, it can be understood that the method can be executed by a vehicle terminal with computing and processing capabilities. Figure 1 The process shown in FIG. 1 is described in detail, and the method for determining the traffic topology map may specifically include the following steps:

[0027] Step 110, obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment.

[0028] As a method, an image of the environment where the vehicle is located can be acquired through an image acquisition device provided on the vehicle. The image acquisition device provided on the vehicle can be a device located at different positions of the vehicle, so that the image of the environment where the vehicle is located acquired is a multi-view image. Optionally, the image of the environment where the vehicle is located can also be a surround view image, and the resolution of the corresponding image can be 1500×2048 or other resolution values, which are not specifically limited here.

[0029] Optionally, the vehicle can be positioned to obtain the vehicle's positioning information, and then the navigation map corresponding to the positioning information can be obtained in the cloud server based on the positioning information. Optionally, it can also be based on the destination entered by the user in the vehicle or an electronic device connected to the vehicle, and a navigation map within a certain range determined by the destination and the vehicle's positioning information can be obtained in the cloud server. Among them, the navigation map (Standard Definition Map, SD Map) mainly stores road-level elements, such as lane lines, etc.

[0030] Step 120, performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by taking the navigation map as prior information and fine-tuning the lane recognition model according to the image.

[0031] As a method, after obtaining the target lane recognition model, the image can be input into the target lane recognition model so that the target lane recognition model can perform lane recognition on the image to obtain a lane recognition result. Optionally, the lane recognition result may include information such as the position information of the lane line, the length of the lane line, and the width of the lane line.

[0032] As a way, when the model is applied offline, it can learn according to the input data. Although some features with good distinguishability can be automatically learned, overfitting features will also be learned, which will reduce the recognition accuracy of the model. Therefore, the recognition accuracy of the model can be improved by adding prior information to the model. In the solution of the present application, the lane recognition model is fine-tuned by adding a navigation map including road-level elements as prior information to the lane recognition model, thereby improving the recognition accuracy of the target lane recognition model.

[0033] Optionally, the navigation map may be input into a lane recognition model so that the lane recognition model is trained according to the navigation map, and the parameters of the lane recognition model are adjusted based on the training results, thereby using the navigation map as prior information.

[0034] Optionally, a priori module may be constructed in the lane recognition model based on the navigation map, so that the features recognized by the lane recognition model are corrected through the navigation map of the seat prior information in the priori module, so as to obtain accurate lane recognition results.

[0035] Optionally, in order to make the lane recognition model closer to the environment where the vehicle is located and to improve the accuracy of lane recognition, the lane recognition model can be fine-tuned by using the acquired image of the environment where the vehicle is located as training data to obtain a target lane recognition model. Optionally, the corresponding lane information in the image can be determined by first matching the image with the navigation map, and then the image can be input into the lane recognition model for training and learning, so that the lane recognition model is closer to the image captured by the vehicle's image acquisition device, thereby improving the recognition accuracy of the lane recognition model.

[0036] Step 130 , performing traffic element recognition on the image using a traffic element recognition model to obtain a traffic element recognition result.

[0037] As a way, in order to enable users to intuitively understand the traffic status of the vehicle's current environment through the traffic topology map and the vehicle computer to control the vehicle's driving in real time according to the traffic topology map, traffic elements can be identified by the images captured by the vehicle's image acquisition device to determine the traffic elements in the vehicle's environment, thereby obtaining the traffic element recognition result.

[0038] Optionally, the traffic element recognition model can be a model including a YOLO network. The backbone network in the traffic element recognition model first extracts features from the image to obtain multiple sets of feature maps with different resolutions and numbers of channels. The multiple feature maps are then input into a feature pyramid network (FNP). The feature pyramid network fuses the feature maps layer by layer in a top-down and lateral connection manner, and outputs feature maps with the same channels but different scales to form a fused feature with a pyramid structure. The fused features are then used through the YOLO network to perform traffic element recognition to obtain traffic element recognition results. Optionally, the traffic element recognition result includes a target box corresponding to the traffic element, a recognition confidence corresponding to the traffic element, and an element category corresponding to the traffic element. Optionally, traffic elements may include elements such as traffic lights, lane direction indicators, and traffic signs. As shown in formula ,in, is the target feature, Identify the traffic elements.

[0039] Step 140 , performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0040] As a method, after obtaining the lane recognition result, a topological analysis can be performed based on the lane lines according to the position information corresponding to different lane lines, so as to determine the positional relationship between different lane lines, and then different lane lines can be connected based on the positional relationship between different lane lines to obtain a traffic topology map. Among them, the positional relationship can indicate the adjacent relationship, the preceding and succeeding relationship, and the guiding relationship between lane lines, etc., which are not specifically limited here.

[0041] Optionally, the traffic topology map may be a lane topology map for indicating the environment in which the vehicle is located, and may display different lanes by lines of different formats. Optionally, the topological relationship in the traffic topology map may indicate the connection relationship and position relationship between different roads.

[0042] As a method, after obtaining the traffic element recognition results and the lane recognition results, the features corresponding to the traffic element recognition results and the lane recognition results are projected into the same space, and then the lane features obtained from the bird's-eye view space and the view conversion matrix from the bird's-eye view to the perspective view are used for matrix multiplication. The transformed lane features and traffic element features are transformed and connected, and then a multi-layer perception network is used to generate a lane-traffic topology prediction based on the lane features and traffic element features after the transformation and connection operations, so as to predict the topological connection between the lane centerlines corresponding to the traffic elements and lane elements, as shown in the formula ,in, Indicates the number of lane lines, Indicates the number of traffic elements, is the lane line information, The traffic element information is then used to generate a traffic topology map through the positional relationship between different lanes and the topological connection between the lane center lines corresponding to the traffic elements and lane elements.

[0043] Optionally, the traffic topology map may include traffic elements corresponding to different lanes, and information such as whether the corresponding lane is currently passable as indicated by the corresponding traffic element.

[0044] In an embodiment of the present application, the target lane recognition model is obtained by using the navigation map corresponding to the environment where the vehicle is located as prior information, and the lane recognition model is fine-tuned by using the image of the environment where the vehicle is located to obtain a target lane recognition model, and then the lane recognition is performed on the image of the environment where the vehicle is located by the target lane recognition model with prior information to obtain a lane recognition result, and the traffic element recognition is performed on the image by the traffic element recognition model to obtain a traffic element recognition result, and finally a topological analysis is performed based on the lane recognition result to obtain a traffic topology map. The present application obtains accurate lane recognition results by using the navigation map with lane information as prior information to recognize the image of the environment where the vehicle is located by the target lane recognition model, and the traffic element recognition results are obtained by recognizing the image of the environment where the vehicle is located by the traffic element recognition model, so that a traffic topology map can be obtained based on the accurate lane recognition results and the traffic element recognition results, thereby enhancing the topological reasoning ability of the vehicle in complex traffic scenes and improving the accuracy and robustness of the traffic topology map.

[0045] See also Figure 2 , Figure 2 The following is a method for determining a traffic topology map provided by an embodiment of the present application. Figure 2 The process shown in FIG. 1 is described in detail, and the method for determining the traffic topology map may specifically include the following steps:

[0046] Step 210, obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment.

[0047] Step 220, performing raster processing on the navigation map to obtain a raster map.

[0048] As a method, grid processing refers to dividing the navigation map into multiple grids. The grid size of the navigation map can be determined first, and then the navigation map is divided into grids based on the grid size to obtain a grid map. Among them, the grid refers to dividing the space into regular grids, and each grid is called a grid. The size of the grid can be determined according to the actual application requirements. For example, for the same space, when the accuracy requirement is high, the space can be divided into a large number of grids, and each grid space is smaller; when the accuracy requirement is low, the space can be divided into a small number of grids, and each grid space is larger. Optionally, the size of the grid can also be related to the resolution and can be set in advance.

[0049] As another method, a grid map pre-set in the cloud server can also be obtained, and the navigation map and the grid map are superimposed to obtain the grid map. Optionally, the navigation map can also be projected onto the grid map to obtain the grid map.

[0050] Step 230: input the grid map into the lane recognition model for reconstruction to obtain a target grid map.

[0051] As a way, the raster map can be reconstructed through the lane recognition model based on the learning method of Masked Autoencoders (MAE), where MAE is a self-supervised learning method that forces the model to learn features in the image by randomly masking part of the input image and then reconstructing the missing pixels.

[0052] Optionally, after the grid map is input into the lane recognition model, the encoder of the lane recognition model can encode the random position of the grid map to randomly add a mask to the grid map, and then decode the masked grid map through the decoder in the lane recognition model, such as Figure 3 As shown, the grid map is reconstructed to obtain the target grid map. Optionally, the grid map can be reconstructed by the following formula: ,in, Reconstructed grid map for the model, For a raster map, Indicates that the masked grid map is encoded by the encoder in the lane recognition model. Indicates that the masked raster map is decoded by the decoder in the lane recognition model.

[0053] Optionally, in the process of decoding the masked grid map through the decoder in the lane recognition model, the pixels of the grid map are reconstructed by re-decoding the masked part to learn the features in the grid map, so that the model can effectively learn the global features of the grid map.

[0054] Optionally, in order to fully learn the characteristics of the grid map, the grid map can be reconstructed multiple times through the MAE learning method to learn the distribution characteristics of the navigation map prior. Each time the reconstruction is completed, the loss function is determined based on the original pixel value of the grid map and the pixel value of the target grid map, and the map reconstruction loss value is calculated according to the loss function. In this way, the parameters of the decoder part corresponding to the lane recognition model can be adjusted according to the map reconstruction loss value, so as to ensure that the navigation map information learned by the lane recognition model is more accurate. Optionally, the formula To determine the map reconstruction loss function, where L is the map reconstruction loss function, N is the set of tokens to which masks are added in the raster map, is the pixel value in the original map, is the pixel value in the reconstructed map. Optionally, when the map reconstruction loss function converges, the step of reconstructing the raster map can be ended; when the map reconstruction loss function does not converge, the parameters of the decoder part corresponding to the lane recognition model can be adjusted according to the map reconstruction loss value until the map reconstruction loss value converges. Optionally, whether the map reconstruction loss function converges can be determined by determining whether the map reconstruction loss value is less than the loss threshold. When the map reconstruction loss value is less than the loss threshold, it is determined that the map reconstruction loss function converges; when the map reconstruction loss value is equal to or greater than the loss threshold, it is determined that the map reconstruction loss function does not converge, or when the number of reconstructions is equal to or greater than the number threshold, it is determined that the map reconstruction loss function converges; when the number of reconstructions is less than the number threshold, it is determined that the map reconstruction loss function does not converge.

[0055] Step 240: fine-tune the lane recognition model based on the image and the target grid map to obtain the target lane recognition model, and perform lane recognition on the image based on the target lane recognition model to obtain a lane recognition result.

[0056] As a way to make the recognition accuracy of the target lane recognition model for subsequent offline applications more accurate, the lane recognition model can be fine-tuned by using the pre-trained model method. Optionally, some parameters can be frozen, and then the unfrozen parameters can be trained and adjusted to obtain new parameters, so as to find the most suitable parameters after multiple trainings to obtain the target lane recognition model.

[0057] Optionally, before fine-tuning, the image and the target grid map may be matched in advance to add corresponding map information to the image, and then the image carrying the map information may be input into the lane recognition model to obtain an initial lane prediction result of the image, and then the loss function of the lane recognition model may be determined based on the initial lane prediction result and the corresponding map information in the image, and some parameters may be randomly selected in the lane recognition model for freezing, so that the unfrozen parameters may be adjusted according to the value of the loss function until the loss function converges, and the target lane recognition model may be obtained based on the updated unfrozen parameters and the frozen parameters.

[0058] Optional, such as Figure 3As shown, the lane recognition model may be used to first perform lane recognition on the image to obtain an initial lane prediction result, and then the initial prediction result may be input into the lane recognition model, and the image may be encoded and decoded according to the encoder in the lane recognition model to correct the initial lane prediction result in the image, and then the encoded and decoded image and the corrected initial lane prediction result may be input into the lane line refinement module of the lane recognition model to perform lane recognition again to obtain the target lane line prediction result. Since the encoder in the lane recognition model is an encoder that is learned by reconstructing the grid map that rasterizes the navigation map, the lane distribution characteristics of the navigation map are learned in the encoder in the lane recognition model, and then the initial lane prediction result may be corrected.

[0059] In some embodiments, Figure 4 As shown, step 240 includes:

[0060] Step 241, adding a mask to the target grid map to obtain a reference grid map.

[0061] As a method, a mask can be randomly added to the target grid map through the MAE encoder to ensure the randomness of the reference grid map. Optionally, in order to improve the accuracy of lane recognition by the lane recognition model, a mask can be added to the grid blocks indicating the lanes in the target grid map to obtain a reference grid map.

[0062] Step 242: perform lane recognition on the image using the lane recognition model to obtain an initial lane recognition result.

[0063] As a way, in order to make the lane recognition model in offline application more compatible with the image collected by the vehicle and its recognition accuracy more accurate, the lane recognition model can be used to perform lane recognition on the image first to obtain the initial lane recognition result, and then the lane recognition model that has learned the map information of the navigation map can be used to correct the initial lane recognition result, so as to adjust the lane recognition model based on the correction result. Optionally, the initial lane recognition result can be information indicating the lane line in the image.

[0064] Step 243: training the lane recognition model using the reference grid map to obtain a reference lane recognition model.

[0065] As a method, the reference grid map can be decoded by the MAE encoder in the lane recognition model to obtain a decoded reference grid map, and the reference grid map and the target grid map are compared to determine the decoding difference, and the parameters of the decoding model in the MAE encoder are adjusted based on the decoding difference, and the reference grid map is decoded again according to the MAE encoder with adjusted parameters until the decoding difference is less than the difference threshold, and the decoding of the reference grid map is terminated, and the reference lane recognition model is obtained based on the MAE encoder with adjusted parameters.

[0066] Step 244 , training the reference lane recognition model according to the initial lane recognition result to obtain the target lane recognition model.

[0067] As a method, after obtaining the initial lane recognition result, the lane recognition difference is determined by the lane information corresponding to the learned navigation map and the lane initial recognition result, and the initial lane recognition result is corrected by the lane information corresponding to the learned navigation map. Then, based on the corrected lane initial recognition result and the lane recognition difference, the parameters of the reference lane recognition model are adjusted until the lane recognition difference indicates that the match degree between the initial lane recognition result and the corresponding lane information in the navigation map is greater than the matching threshold, and the training of the reference lane recognition model is terminated to obtain the target lane recognition model.

[0068] Please continue reading Figure 2 , step 250, performing traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result.

[0069] Step 260 , performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0070] The specific step descriptions of step 210 and step 250 - step 260 can refer to step 110 and step 130 - step 140, which will not be repeated here.

[0071] In this embodiment, a raster map can be obtained by performing raster mapping on the navigation map, and then the raster map is input into the lane recognition model for reconstruction to obtain a target raster map, and then a reference raster map is obtained by adding a mask to the target raster map, and the lane recognition model is used to perform lane recognition on the image to obtain an initial lane recognition result, and then the lane recognition model is trained based on the reference raster map and the initial lane recognition result, so that the target lane recognition model can learn the lane information in the navigation map as prior information, and the target lane recognition model can be more adapted to the current environment of the vehicle, thereby improving the recognition accuracy of the target lane recognition model, and then improving the accuracy of the traffic topology map.

[0072] See also Figure 5 , Figure 5 The following is a method for determining a traffic topology map provided by an embodiment of the present application. Figure 5 The process shown in FIG. 1 is described in detail, and the method for determining the traffic topology map may specifically include the following steps:

[0073] Step 310, obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment.

[0074] Step 320, performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by taking the navigation map as prior information and fine-tuning the lane recognition model according to the image.

[0075] The specific step descriptions of step 310 to step 320 may refer to step 110 to step 120, which will not be repeated here.

[0076] Step 330 performs feature extraction on the image to obtain multi-scale image features, and determines target features corresponding to the foreground view of the vehicle in the multi-scale image features.

[0077] As a method, features of the image can be extracted by ResNet-50, VOV and Swin-B as feature extraction networks, so as to obtain multiple sets of image features with different resolutions and channel numbers, and then the target features corresponding to the foreground view of the vehicle are determined in the multi-scale image features. Optionally, in the process of the vehicle's formation, the traffic elements located in the front view of the vehicle play a vital role in the driving of the vehicle, so the foreground view of the vehicle can be used as the target feature in the multi-scale image features to identify the traffic elements.

[0078] Step 340 , performing traffic element recognition on the target feature based on the traffic element recognition model to obtain feature vectors of instances corresponding to multiple traffic elements.

[0079] As a method, YOLO in the traffic element recognition model can be used to identify traffic elements on target features, so as to obtain feature vectors of instances corresponding to multiple traffic elements.

[0080] Step 350 : determining a plurality of traffic element target frames and the traffic element information corresponding to each of the plurality of traffic element target frames according to the feature vector.

[0081] As a method, after obtaining feature vectors of instances corresponding to multiple traffic elements, the feature vectors can be decoded by a decoder in a traffic element recognition model to obtain information such as bounding boxes corresponding to each of the multiple traffic elements, confidence levels corresponding to each of the multiple traffic elements, and categories corresponding to each of the multiple traffic elements, and the bounding boxes corresponding to each of the multiple traffic elements are determined as target boxes corresponding to each of the multiple traffic elements, and the confidence levels and categories corresponding to each of the multiple traffic elements are determined as traffic element information corresponding to each of the multiple traffic element target boxes.

[0082] Step 360: Determine the traffic element recognition result according to the multiple traffic element target frames and the traffic element information.

[0083] As one approach, multiple traffic element target frames and traffic element information are associated to obtain an association relationship between the traffic element target frames and the traffic element information, thereby determining the association relationship as a traffic element recognition result.

[0084] Please continue reading Figure 5 , step 370, performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain the traffic topology map.

[0085] In some embodiments, Figure 6 As shown, step 370 includes:

[0086] Step 371 , determining traffic element information corresponding to the environment according to the traffic element recognition result, and determining lane line information corresponding to the environment according to the lane recognition result.

[0087] As a method, after obtaining the traffic element recognition result, the traffic element information related to the form of the vehicle in the traffic element recognition result can be determined first, such as the traffic light type, the current drivable direction indicated by the traffic light, and the lane marking (traffic element markings indicating that the lane is a straight lane, a circular lane, a right turn lane, a left turn lane, a U-turn lane, and a merging lane), so as to facilitate the subsequent determination of the topological relationship between the traffic element and the lane line. Optionally, the lane line information corresponding to the environment where the vehicle is located can be determined by the lane markings used to indicate different lanes in the lane recognition result, wherein the lane line information may include the color of the lane line, the type of the lane line (such as double yellow line, solid line, and dashed line, etc.), and the position of the lane line, etc.

[0088] Step 372: Perform a topological analysis based on the traffic element information and the lane line information to determine the topological relationship between the traffic element and the lane line.

[0089] As a method, after obtaining traffic element information and lane line information, the traffic element corresponding to each lane line can be determined, and the lane center lines corresponding to different lanes can be determined based on multiple lane lines. Then, the topological relationship between the lane lines of the Giti Yuansun diagram can be determined based on the traffic element corresponding to each lane line and the corresponding lane center line.

[0090] In some embodiments, step 372 includes: determining the positional relationship between the multiple lane lines based on the lane line information, and determining the topological connection between different lanes based on the positional relationship; and determining the topological relationship based on the topological connection and the traffic element information.

[0091] As a method, lane lines belonging to the same lane and lane lines that are the same lane can be determined based on lane line information, and then the adjacent relationship and connection relationship between lanes can be determined based on the lane lines of the same lane and the lane lines that are the same lane. Information belonging to the drainage lane line can be determined based on the lane line information, and the connection relationship between lanes can be determined based on the lane connected to the drainage lane line, so as to determine the topological connection between different lanes based on the adjacent relationship and the connection relationship.

[0092] Optionally, after determining the topological connection between different lanes, the traffic element features in the traffic element information and the lane features in the lane information can be projected into the same space, and then the lane features obtained from the bird's-eye view space and the view conversion matrix from the bird's-eye view to the perspective view are used to perform a matrix multiplication operation. The transformed lane features and traffic element features are transformed and connected, and then a multi-layer perception network is used to generate a lane-traffic topological relationship based on the lane features and traffic element features after the transformation and connection operations, and then the topological relationship is determined based on the lane-traffic topological relationship and the topological connection between different lanes.

[0093] Step 373: determine the traffic topology map according to the topological relationship.

[0094] As a way, the topological relationship includes the topological connection between different lanes and the topological relationship between traffic elements and lanes. Therefore, the generated traffic topology map can clearly express the positional relationship of lane center lines corresponding to different lanes, the traffic elements corresponding to different lanes, and the traffic information indicated by each traffic element.

[0095] In this embodiment, multi-scale image features can be obtained by first performing feature extraction on the image, so that the target features corresponding to the foreground view of the vehicle can be determined in the multi-scale image features, so that traffic elements can be recognized on the target features through the traffic element recognition model, and feature vectors of instances corresponding to each traffic element can be obtained, thereby improving the recognition accuracy of the feature vectors, and then multiple traffic element target frames and traffic element information corresponding to each of the multiple traffic element target frames can be determined through the feature vectors, so that the traffic element recognition results can be determined based on the multiple traffic element target frames and the traffic element information, thereby ensuring the accuracy of the traffic element recognition results, and then topological analysis is performed on the traffic element recognition results and the lane recognition results to obtain a traffic topology map, thereby improving the richness and accuracy of the traffic topology map.

[0096] See also Figure 7 , Figure 7 The following is a method for determining a traffic topology map provided by an embodiment of the present application. Figure 7 The process shown in FIG. 1 is described in detail, and the method for determining the traffic topology map may specifically include the following steps:

[0097] Step 410, obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment.

[0098] Step 420: extract features from the image to obtain multi-scale image features, and perform feature fusion based on the multi-scale image features to obtain fused features.

[0099] As a method, feature extraction networks can be used to extract features from images to obtain multiple sets of image features with different resolutions and numbers of channels. For example, ResNet-50, VOV, and Swin-B are used to extract image features to generate four sets of feature maps with different resolutions and numbers of channels. , , , }, where the resolution of the feature map corresponds to 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the input image, and the number of channels increases successively.

[0100] Optionally, after obtaining the multi-scale image features, the feature maps of different resolutions and numbers of channels can be input into the feature fusion network for feature fusion to obtain fused features. Optionally, feature fusion can be performed through a feature pyramid network, which can fuse the feature maps of the above groups of different resolutions and numbers of channels layer by layer in a top-down and horizontally connected manner, and output feature maps with the same channels but different scales to form a fused feature with a pyramid structure.

[0101] Step 430: Perform visual conversion based on the fusion features to obtain bird's-eye view image features.

[0102] As a method, after obtaining the fused features, in order to obtain the complete lane line information around the lane, the fused features can be visually converted first, so that the fused features are converted into bird's-eye view features, and then the obtained lane recognition results can fully indicate the lane line information of the vehicle's environment.

[0103] Step 440, inputting the bird's-eye view image features into the target lane recognition model for lane recognition to obtain the lane recognition result, wherein the target lane recognition model is obtained by using the navigation map as prior information and fine-tuning the lane recognition model according to the image.

[0104] As a method, the target lane recognition model can be used to perform lane recognition on the features of the bird's-eye view image to obtain a reference lane recognition result, and the reference lane recognition result is input again into the module including the prior information in the target lane recognition model for correction, so as to obtain the lane recognition result. Optionally, the lane recognition result may include information such as the position information of the lane line, the length of the lane line, and the width of the lane line.

[0105] Step 450 , performing traffic element recognition on the image using a traffic element recognition model to obtain a traffic element recognition result.

[0106] Step 460 , performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0107] The specific step descriptions of step 410 and step 450 - step 460 can refer to step 110 and step 130 - step 140, which will not be repeated here.

[0108] In this embodiment, when lane recognition is performed on an image through a target lane recognition model, feature extraction may be performed on the image first to obtain multi-scale image features, and the multi-scale image features may be fused to obtain fused features, and then the fused features may be visually transformed to obtain bird's-eye view image features, so that the target lane recognition model can perform lane recognition on the bird's-eye view image features, thereby improving the comprehensiveness and accuracy of the lane recognition results, thereby ensuring the accuracy of the traffic topology map.

[0109] See also Figure 8 , Figure 8 The following is a method for determining a traffic topology map provided by an embodiment of the present application. Figure 8 The process shown in FIG. 1 is described in detail, and the method for determining the traffic topology map may specifically include the following steps:

[0110] Step 510, obtaining an image of the environment in which the vehicle is located and a navigation map corresponding to the environment.

[0111] Step 520, performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by taking the navigation map as prior information and fine-tuning the lane recognition model according to the image.

[0112] Step 530 , performing traffic element recognition on the image using a traffic element recognition model to obtain a traffic element recognition result.

[0113] Step 540 , performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0114] The specific step descriptions of step 510 to step 540 may refer to step 110 to step 140 and will not be repeated here.

[0115] Step 550, determining the target lane where the vehicle is located.

[0116] As a method, after obtaining the traffic topology map, the vehicle can be positioned first to obtain the vehicle positioning information, and then the target lane where the vehicle is located can be determined in the traffic topology map according to the vehicle positioning information. Optionally, the target lane where the lane is currently located can also be determined by identifying the image collected by the vehicle.

[0117] Step 560: Obtain navigation information of the vehicle, and determine the traffic information of the vehicle in the traffic topology map according to the navigation information and the target lane.

[0118] As a method, the navigation information of the vehicle may include the driving lane of the vehicle at the current intersection or the next intersection, so that the traffic information of the vehicle can be determined in the traffic topology map based on the navigation information.

[0119] Optionally, if the current lane is determined to be impassable and different from the lane indicated by the navigation information based on the navigation information and the target lane in the traffic topology map, the target traffic information corresponding to the lane consistent with the navigation information in the traffic topology map can be used, and the vehicle's traffic information can be determined based on the target traffic information; if the current lane is determined to be impassable and consistent with the lane indicated by the navigation information based on the navigation information and the target lane in the traffic topology map, the target traffic information corresponding to the target lane in the traffic topology map can be used, and the vehicle's traffic information can be determined based on the target traffic information. Optionally, the traffic information includes information such as the current lane is impassable, the navigation lane is impassable, the navigation lane is impassable, and the current lane is impassable.

[0120] Step 570: Provide passage instructions to the vehicle according to the passage information.

[0121] As a method, after obtaining the vehicle's traffic information, a vehicle traffic prompt can be generated based on the vehicle's traffic information, and the communication prompt can be used as a traffic indication. Optionally, the traffic indication can include a lane change indication, a straight indication, a right turn indication, a left turn indication, and a waiting traffic light indicator.

[0122] In this embodiment, after the traffic topology map is determined, the vehicle's traffic information is determined in the traffic topology map by obtaining the vehicle's navigation information and determining the target lane where the vehicle is located, so that the vehicle is instructed to pass based on the traffic information to ensure the vehicle's driving safety.

[0123] Fig. 9 FIG. 1 is a flow chart of a method for determining a traffic topology map according to an embodiment of the present application. Fig. 9 As shown in the figure, when the vehicle is powered on and enters the autonomous driving mode, the vehicle's camera is exposed, the video stream is input, and transmitted to the on-board computing unit through the interface. A single frame image of the video stream is extracted, and after preprocessing, resizing, and data normalization steps, it is input into the lane recognition model and the traffic element recognition model for inference calculation; the traffic element recognition model downsamples the input image through the skeleton network to generate feature maps with different resolutions and numbers of channels, and this feature map is input into the feature pyramid network to output multiple feature maps with the same channels but different scales.

[0124] The lane recognition model is trained based on the navigation map using the MAE learning method to learn and understand the distribution information of the navigation map. First, the MAE is trained using self-supervised learning to obtain the distribution of the navigation map on the pre-training dataset. During training, the navigation map is first processed into a segmentation map, and then a mask is randomly added to the segmentation map. The encoder is used to encode the segmentation map with the mask added to restore the map semantic segmentation map. Therefore, a layer of semantic segmentation head is added to the lane recognition model to obtain the final semantic segmentation result. Secondly, the module is fine-tuned through real-time acquired images, and the lane recognition model is used to make an initial prediction of lane recognition on the image as input. The pre-trained MAE module is used to enhance the result so that the initial prediction result can be more consistent with the actual distribution of the navigation map.

[0125] The trained lane recognition model then performs lane recognition on the multi-scale image features to obtain initial predictions of generated lane elements. These initial predictions are then fed into the prior module in the fine-tuned lane recognition model and processed through the final lane segmentation head (i.e., the prior module including the prior information) to obtain accurate lane elements.

[0126] Then, the traffic element recognition model uses the YOLO detector, takes the forward-looking multi-scale image features in the multi-scale image features as input, obtains the feature vector corresponding to the instance of the traffic element, and then decodes the feature vector to output the confidence, category and bounding box of the corresponding traffic element. Among them, traffic elements include but are not limited to traffic lights, speed limit signs, etc.

[0127] Then, the topological relationship between traffic elements and lane line instances of different categories is analyzed through the traffic element-lane line topology branch. Both traffic element and lane line features are projected into the same space. Matrix multiplication is performed using the lane features obtained from the bird's-eye view space and the view transformation matrix from the bird's-eye view to the perspective view. The converted lane features and traffic element features are transformed and concatenated, and then a multi-layer perception network is used to generate lane-traffic topology predictions.

[0128] Finally, by analyzing the positional relationship between different lane line instances, the topological structure between the predicted lane center lines is determined, and then a complete lane topology map is generated based on the peeling structure and the traffic element-lane line topological branch. The positional relationship between different lane line instances includes the connection and relative relationship between lane lines.

[0129] Fig.10 is a block diagram of a device for determining a traffic topology map according to an embodiment of the present application, such as Fig.10 As shown, the traffic topology map determination device 600 includes: a navigation map acquisition module 610, a lane recognition module 620, a traffic element recognition module 630 and a topology analysis module 640.

[0130] A navigation map acquisition module 610 is used to acquire an image of the environment in which the vehicle is located and a navigation map corresponding to the environment; a lane recognition module 620 is used to perform lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by using the navigation map as prior information and fine-tuning the lane recognition model according to the image; a traffic element recognition module 630 is used to perform traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result; a topology analysis module 640 is used to perform topology analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map.

[0131] In some embodiments, the lane recognition module 620 includes: a grid submodule, used to perform grid processing on the navigation map to obtain a grid map; a target grid map determination submodule, used to input the grid map into the lane recognition model for reconstruction to obtain a target grid map; an adjustment submodule, used to fine-tune the lane recognition model based on the image and the target grid map to obtain the target lane recognition model.

[0132] In some embodiments, the adjustment submodule includes: a reference grid map determination unit, used to add a mask to the target grid map to obtain a reference grid map; a lane recognition unit, used to perform lane recognition on the image through the lane recognition model to obtain an initial lane recognition result; a first training unit, used to train the lane recognition model through the reference grid map to obtain a reference lane recognition model; and a second training unit, used to train the reference lane recognition model according to the initial lane recognition result to obtain the target lane recognition model.

[0133] In some embodiments, the topology analysis module 640 includes: an information determination submodule, used to determine the traffic element information corresponding to the environment based on the traffic element recognition result, and determine the lane line information corresponding to the environment based on the lane recognition result; a topological relationship determination submodule, used to perform topological analysis based on the traffic element information and the lane line information, and determine the topological relationship between the traffic elements and the lane lines; a traffic topology map determination submodule, used to determine the traffic topology map based on the topological relationship.

[0134] In some embodiments, the topological relationship determination submodule includes: a topological connection determination unit, used to determine the positional relationship between the multiple lane lines based on the lane line information, and determine the topological connection between different lanes based on the positional relationship; a topological relationship determination unit, used to determine the topological relationship based on the topological connection and the traffic element information.

[0135] In some embodiments, the traffic element recognition module 630 includes: a target feature determination submodule, which is used to extract features from the image to obtain multi-scale image features, and determine the target features corresponding to the foreground view of the vehicle in the multi-scale image features; a feature vector determination submodule, which is used to perform traffic element recognition on the target features based on the traffic element recognition model to obtain feature vectors of multiple traffic element corresponding instances; a traffic element information determination submodule, which is used to determine multiple traffic element target boxes and the traffic element information corresponding to each of the multiple traffic element target boxes according to the feature vectors; and a traffic element recognition result determination submodule, which is used to determine the traffic element recognition result according to the multiple traffic element target boxes and the traffic element information.

[0136] In some embodiments, the lane recognition module 620 includes: a feature fusion submodule, which is used to extract features from the image to obtain multi-scale image features, and perform feature fusion based on the multi-scale image features to obtain fusion features; a bird's-eye view image feature determination submodule, which is used to perform visual conversion based on the fusion features to obtain bird's-eye view image features; and a lane recognition result determination submodule, which is used to input the bird's-eye view image features into the target lane recognition model to perform lane recognition and obtain the lane recognition result.

[0137] According to one aspect of the embodiments of the present application, an electronic device is also provided, such as Fig.11 As shown, the electronic device 700 includes a processor 710 and one or more memories 720. The one or more memories 720 are used to store program instructions executed by the processor 710. When the processor 710 executes the program instructions, the above-mentioned method for determining the traffic topology map is implemented.

[0138] Furthermore, the processor 710 may include one or more processing cores. The processor 710 runs or executes instructions, programs, code sets or instruction sets stored in the memory 720, and calls data stored in the memory 720. Optionally, the processor 710 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 710 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It is understandable that the above-mentioned modem may not be integrated into the processor, but may be implemented separately through a communication chip.

[0139] According to one aspect of the present application, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or may exist independently without being assembled into the electronic device. The above computer-readable storage medium carries computer-readable instructions, and when the computer-readable storage instructions are executed by a processor, the method in any of the above embodiments is implemented.

[0140] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0141] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0142] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0143] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0144] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for determining a traffic topology map, characterized in that: The method comprises: Acquire an image of the environment in which the vehicle is located and a navigation map corresponding to the environment; Performing lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by taking the navigation map as prior information and fine-tuning the lane recognition model according to the image; Performing traffic element recognition on the image using a traffic element recognition model to obtain a traffic element recognition result; Performing topological analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map; The target lane recognition model is specifically obtained in the following manner: Performing raster processing on the navigation map to obtain a raster map; Inputting the grid map into the lane recognition model for reconstruction to obtain a target grid map; Adding a mask to the target grid map to obtain a reference grid map; Performing lane recognition on the image using the lane recognition model to obtain an initial lane recognition result; Training the lane recognition model through the reference grid map to obtain a reference lane recognition model; The reference lane recognition model is trained according to the initial lane recognition result to obtain the target lane recognition model.

2. The method according to claim 1, characterized in that The performing topological analysis according to the traffic element recognition result and the lane recognition result to obtain the traffic topology map includes: Determining traffic element information corresponding to the environment according to the traffic element recognition result, and determining lane line information corresponding to the environment according to the lane recognition result; Performing a topological analysis based on the traffic element information and the lane line information to determine a topological relationship between the traffic element and the lane line; The traffic topology map is determined according to the topological relationship.

3. The method according to claim 2, characterized in that The performing topological analysis based on the traffic element information and the lane line information to determine the topological relationship between the traffic element and the lane line includes: Determining a positional relationship between a plurality of lane lines based on the lane line information, and determining a topological connection between different lanes based on the positional relationship; The topological relationship is determined according to the topological connection and the traffic element information.

4. The method according to claim 1, characterized in that: The step of performing traffic element recognition on the image using a traffic element recognition model to obtain a traffic element recognition result includes: Extracting features from the image to obtain multi-scale image features, and determining target features corresponding to the foreground view of the vehicle in the multi-scale image features; Based on the traffic element recognition model, the target feature is subjected to traffic element recognition to obtain feature vectors of instances corresponding to multiple traffic elements; Determine a plurality of traffic element target frames and the traffic element information corresponding to each of the plurality of traffic element target frames according to the feature vector; The traffic element recognition result is determined according to the multiple traffic element target frames and the traffic element information.

5. The method according to claim 1, characterized in that The performing lane recognition on the image based on the target lane recognition model to obtain a lane recognition result includes: Extracting features from the image to obtain multi-scale image features, and fusing features based on the multi-scale image features to obtain fused features; Perform visual conversion based on the fusion features to obtain bird's-eye view image features; The bird's-eye view image features are input into the target lane recognition model to perform lane recognition to obtain the lane recognition result.

6. A device for determining a traffic topology map, characterized in that: The device comprises: A navigation map acquisition module, used to acquire an image of the environment in which the vehicle is located and a navigation map corresponding to the environment; A lane recognition module, used to perform lane recognition on the image based on a target lane recognition model to obtain a lane recognition result, wherein the target lane recognition model is obtained by taking the navigation map as prior information and fine-tuning the lane recognition model according to the image; A traffic element recognition module, used to perform traffic element recognition on the image through a traffic element recognition model to obtain a traffic element recognition result; A topology analysis module, used to perform topology analysis based on the traffic element recognition result and the lane recognition result to obtain a traffic topology map; Among them, the traffic topology map determination device is used to perform grid processing on the navigation map to obtain a grid map; input the grid map into the lane recognition model for reconstruction to obtain a target grid map; add a mask to the target grid map to obtain a reference grid map; perform lane recognition on the image through the lane recognition model to obtain an initial lane recognition result; train the lane recognition model through the reference grid map to obtain a reference lane recognition model; train the reference lane recognition model according to the initial lane recognition result to obtain the target lane recognition model.

7. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 5.

Images