A vehicle positioning method, device, apparatus and storage medium

Abstract

A vehicle positioning method, device, equipment and storage medium are provided, which comprise: acquiring multiple target vehicle surround view images; splicing the multiple target vehicle surround view images to obtain multiple target vehicle surround view spliced images; performing semantic segmentation on each surround view spliced image to obtain a vehicle segmented image; converting the segmented image into a single-channel image according to a semantic category and performing rasterization processing to obtain a vehicle raster image of different semantic categories; aligning and splicing the vehicle raster images of two different semantic categories based on the pose transformation relationship of the vehicle to obtain a local semantic raster image; extracting multiple target vehicle semantic features of the local semantic raster image to obtain a semantic feature sequence; performing similarity matching based on the semantic feature sequence and a semantic feature sequence of a preset global map to determine multiple candidate road segments; and matching point cloud data of the segmented image of the target position with candidate point cloud data of each candidate road segment to position the target vehicle.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, specifically to a vehicle positioning method, device, equipment, and storage medium. Background Technology

[0002] With the rapid development of science and technology, people are increasingly pursuing convenience and speed, hence the emergence of autonomous driving technology. However, in the field of autonomous driving, accurate vehicle positioning is particularly important. In normal environments, we locate vehicles based on satellite positioning (GPS) signals.

[0003] However, in real-world driving, vehicles not only operate in environments with clear GPS signals but also in complex environments with relatively weak GPS information. When autonomous vehicles are in situations where accurate GPS positioning information is unavailable, they need to use additional sensors to collect environmental information and then match it with a pre-created map to achieve positioning. However, since typical visual algorithms extract local regions with specific pixel distribution characteristics from images and then calculate features based on changes in pixel distribution, this method is easily affected by lighting or noise, causing the features to lack robustness over long periods and change over time, leading to a decrease in matching success rate. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the present invention provides a vehicle positioning method to solve the above technical problems.

[0005] This invention provides a vehicle localization method, the method comprising: acquiring multiple surround view images of a target vehicle at multiple vehicle acquisition locations along the driving path of a target vehicle; stitching the multiple surround view images of the target vehicle based on the vehicle acquisition locations to obtain a target vehicle surround view stitched image at multiple vehicle acquisition locations; performing semantic segmentation on each of the target vehicle surround view stitched images to obtain a vehicle segmentation image; converting the vehicle segmentation image into a single-channel image according to the semantic category obtained by semantic segmentation, and performing rasterization processing to obtain vehicle raster images of different semantic categories; and based on the pose transformation relationship between two consecutive vehicle acquisition locations, [the method further involves] [the method involving ... Vehicle raster images of different semantic categories are aligned and stitched together until all vehicle raster images are stitched together to obtain the target vehicle semantic raster image of the target vehicle; multiple target vehicle semantic features are extracted from the target vehicle semantic raster image to obtain a target vehicle semantic feature sequence; similarity matching is performed between the target vehicle semantic feature sequence and the global semantic feature sequence of a preset global map to determine multiple candidate road segments, where each candidate road segment is a road segment in the preset global map; the point cloud data of the segmented image of the target vehicle's acquisition location is matched with the candidate point cloud data of each candidate road segment to locate the target vehicle.

[0006] In one embodiment of the present invention, before performing similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of a preset global map, the method further includes: acquiring multiple target area surround view images from multiple area acquisition locations in multiple road segments of the target area, wherein the target area includes the driving path; stitching the multiple target area surround view images based on the area acquisition locations to obtain a target area surround view stitched image from multiple area acquisition locations; performing semantic segmentation on each target area surround view stitched image to obtain a region segmentation image; converting the segmented image into a single-channel image according to the semantic category obtained by semantic segmentation and performing rasterization processing to obtain region raster images of different semantic categories; aligning and stitching the region raster images of different semantic categories from the two consecutive area acquisition locations based on the pose transformation relationship between them, until all the region raster images are stitched together to obtain a target area local semantic raster image of the target area; and extracting multiple target area semantic features from the target area local semantic raster image to obtain a global semantic feature sequence of the preset global map.

[0007] In one embodiment of the present invention, stitching multiple target vehicle surround view images based on the vehicle acquisition location to obtain a target vehicle surround view stitched image at the vehicle acquisition location includes: acquiring multiple target vehicle surround view images using a camera, and performing image distortion correction on the target vehicle surround view images based on the extrinsic and intrinsic parameters of the camera to obtain a distortion-free target vehicle surround view image, wherein the multiple target vehicle surround view images include at least one of the front image, rear image, left image, and right image of the target vehicle; projecting the distortion-free target vehicle surround view image onto the ground through an affine transformation to obtain a road top view image; and stitching the road top view image based on the extrinsic and intrinsic parameters of the camera to obtain the target vehicle surround view stitched image.

[0008] In one embodiment of the present invention, semantic segmentation is performed on each of the target vehicle surround-view stitched images to obtain vehicle segmented images. The vehicle segmented images are then converted into single-channel images according to the semantic categories obtained from the semantic segmentation and rasterized to obtain vehicle raster images with different semantic categories. This includes: acquiring a target vehicle surround-view stitched image, the target vehicle surround-view stitched image including multiple semantic features; inputting the target vehicle surround-view stitched image into a semantic recognition model, and segmenting the target vehicle surround-view stitched image into multiple vehicle segmented images according to the semantic features; converting the multiple vehicle segmented images into single-channel images according to the semantic feature categories to obtain multiple target vehicle single-channel images with the same number of semantic features; and downsampling and rasterizing the multiple target vehicle single-channel images to obtain raster images with different semantics.

[0009] In one embodiment of the present invention, determining multiple candidate road segments by performing similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of a preset global map includes: the preset global map includes multiple semantic features, multiple trajectory points, and lane lines of the target vehicle, wherein the multiple trajectory points are multiple fixed points on the driving trajectory of the target vehicle; the preset global map is divided into multiple preset global map local maps based on the lane lines of the target vehicle and the multiple trajectory points; each preset global map local map is segmented into multiple preset global map local semantic maps based on the semantic features of the preset global map, and multiple preset global map local single-channel images are obtained based on the multiple preset global map local semantic maps; the multiple global map local single-channel images are rasterized to obtain multiple preset global map local semantic raster images.

[0010] In one embodiment of the present invention, determining multiple candidate road segments by performing similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of a preset global map further includes: extracting the image distance of the multiple single-channel images, and determining the feature vector of the local semantic raster of the preset global map based on the image distance and channel order; obtaining a semantic raster of the target vehicle's driving road based on the target vehicle's driving trajectory, and extracting the feature vector of the semantic raster of the target vehicle's driving road; matching the feature vector of the local semantic raster of the preset global map and the feature vector of the semantic raster of the target vehicle's driving road item by item based on different semantics to obtain an image distance value; when the image distance value is less than a preset threshold, it is determined that the match is successful, and multiple first candidate road segments are obtained.

[0011] In one embodiment of the present invention, matching the point cloud data of the segmented image based on the target vehicle's acquisition location with the candidate point cloud data of each of the candidate road segments to locate the target vehicle includes: obtaining a preliminary position at the moment the target vehicle's current position moves down based on the target vehicle's driving trajectory; obtaining a feature vector of the preliminary position semantic raster image based on the preliminary position, and matching the feature vector of the preliminary position semantic raster image with the feature vector of a local semantic raster image of the complete road map one by one based on different semantics to obtain multiple second candidate road segments; comparing the relative relationship between the target vehicle's current position and the preliminary position, and the relative positional relationship between the multiple first candidate road segments and the multiple second candidate road segments, if the relative positional relationship between any first candidate road segment and any second candidate road segment is the same as the relative relationship between the target vehicle's current position and the preliminary position, then determining that the road segment formed by the first road segment and the second road segment is the road segment where the target vehicle is located.

[0012] In one embodiment of the present invention, after determining the road segment where the target vehicle is located, the method further includes: acquiring lane line information on both sides of the road segment where the target vehicle is located, and determining the target vehicle and the lane line information on both sides of the road segment where the target vehicle is located based on the lane line information on both sides of the road segment where the target vehicle is located, and determining the lateral coordinate relationship between the target vehicle body and the lane; fusing the lateral coordinate relationship between the target vehicle body and the lane with the road segment information where the target vehicle is located to obtain the target vehicle positioning fusion information; and performing rasterization processing on the target vehicle positioning fusion information to obtain the positioning of the target vehicle.

[0013] This invention provides a vehicle positioning device, comprising: an image acquisition module for acquiring multiple surround view images of a target vehicle at multiple vehicle acquisition locations along the driving path of a target vehicle; a first image stitching module for stitching the multiple surround view images of the target vehicle based on the vehicle acquisition locations to obtain a surround view stitched image of the target vehicle at multiple vehicle acquisition locations; an image segmentation module for performing semantic segmentation on each of the surround view stitched images of the target vehicle to obtain a segmented vehicle image; an image conversion module for converting the segmented vehicle image into a single-channel image according to the semantic category obtained by semantic segmentation, and performing rasterization processing to obtain vehicle raster images of different semantic categories; and a second image stitching module for stitching the images based on the pose transformation relationship between the preceding and following vehicle acquisition locations. The vehicle raster images of different semantic categories at the latter two vehicle acquisition locations are aligned and stitched together until all vehicle raster images are stitched together to obtain the target vehicle local semantic raster image; the feature extraction module is used to extract multiple target vehicle semantic features from the target vehicle local semantic raster image to obtain a target vehicle semantic feature sequence; the feature matching module is used to perform similarity matching based on the target vehicle semantic feature sequence and the global semantic feature sequence of the preset global map to determine multiple candidate road segments, wherein the candidate road segment is one of the road segments in each preset global map; the vehicle localization module is used to match the point cloud data of the segmented image of the target location with the candidate point cloud data of each candidate road segment to locate the target vehicle.

[0014] The present invention provides an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device enables the vehicle positioning method described above.

[0015] The present invention provides a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed by a computer processor, causes the computer to perform the vehicle positioning method as described above.

[0016] The beneficial effects of this invention are as follows: The vehicle localization method, apparatus, device, and storage medium proposed in this invention acquire vehicle surround view images, perform semantic segmentation and rasterization processing on the vehicle surround view images to obtain different semantic raster images, and then obtain the corresponding feature vectors based on the semantic raster images. The obtained vehicle semantic raster image feature vectors are compared with the global map feature vectors to obtain a coarse localization of the vehicle. This method can obtain the localization of the target vehicle based on an onboard camera without requiring additional external equipment. Furthermore, the single-channel image conversion and rasterization processing, based on the semantic features of the image, determines the vehicle localization to a certain extent, reducing interference from lighting or noise and enhancing the robustness of the data.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0019] Figure 1 This is a schematic diagram illustrating the implementation environment of a vehicle positioning method according to an exemplary embodiment of this application;

[0020] Figure 2 This is an exemplary embodiment of the present application illustrating a flowchart of a vehicle positioning method;

[0021] Figure 3 This is a schematic diagram illustrating a ground visual semantic mapping method in an exemplary embodiment of this application;

[0022] Figure 4 This is a schematic diagram illustrating a ground-based visual semantic localization method as an exemplary embodiment of this application;

[0023] Figure 5 This is a schematic diagram illustrating feature calculation based on semantic map tiles, as shown in an exemplary embodiment of this application.

[0024] Figure 6 This is a schematic diagram illustrating a slice-based semantic feature sequence as an exemplary embodiment of this application;

[0025] Figure 7 This is a block diagram illustrating a vehicle positioning device in another exemplary embodiment of this application;

[0026] Figure 8A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0027] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0028] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0029] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0030] First, it's important to clarify that image moments (HU moments) are linear combinations of normalized moments. HU moments remain invariant after image rotation, translation, scaling, and other operations, so they are frequently used to identify image features. Hidden Markov Models (HMMs) are statistical models used to describe a Markov process with hidden, unknown parameters.

[0031] Figure 1 This is a schematic diagram illustrating the implementation environment of a vehicle positioning method according to an exemplary embodiment of this application.

[0032] Figure 1 This is a schematic diagram illustrating the implementation environment of an automatic emergency braking method for a vehicle, as shown in an exemplary embodiment of this application. Figure 1As shown, the system architecture may include the current vehicle 101 and computer device 102. The computer device 102 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, or a neural network computer. Those skilled in the art can use the computer device 102 to create a complete road map based on surround-view images acquired by a camera, and then match the complete road map with local images surrounding the target vehicle to obtain the target vehicle's location information.

[0033] Figure 2 This is an exemplary embodiment of the present application illustrating a vehicle positioning method flowchart.

[0034] like Figure 2 As shown, in an exemplary embodiment, the vehicle positioning method includes at least steps S210 to S280, which are described in detail below:

[0035] Step S210: Obtain multiple surround view images of the target vehicle at multiple vehicle collection locations along the target vehicle's driving path.

[0036] It should be understood that the vehicle itself is equipped with many sensors, including but not limited to an onboard camera, which is always activated while the vehicle is in motion and can collect and save environmental image information during the vehicle's journey.

[0037] During the vehicle's movement, multiple locations can be selected as data acquisition points. The vehicle's onboard cameras capture surround-view images of the target vehicle. The target vehicle may be equipped with one or more onboard cameras. The shooting angle of one camera can be adjusted to obtain multiple surround-view images of the target vehicle, including information about its surroundings. Alternatively, at least two onboard cameras positioned around the vehicle can each capture a surround-view image. Ideally, the multiple surround-view images should encompass 360-degree image information of the target vehicle.

[0038] Step S220: Stitch together multiple target vehicle surround view images based on the vehicle acquisition locations to obtain a target vehicle surround view stitched image from multiple vehicle acquisition locations.

[0039] Stitching multiple surround view images of a target vehicle based on the vehicle's acquisition location to obtain a stitched surround view image of the target vehicle at that location includes: acquiring multiple surround view images of the target vehicle using a camera, and performing image distortion correction on the surround view images of the target vehicle based on the camera's extrinsic and intrinsic parameters to obtain a distortion-free surround view image of the target vehicle. These multiple surround view images of the target vehicle include at least one of the front, rear, left, and right images of the target vehicle; projecting the distortion-free surround view image of the target vehicle onto the ground through an affine transformation to obtain a road top-view image; and stitching the road top-view image based on the camera's extrinsic and intrinsic parameters to obtain a stitched surround view image of the target vehicle.

[0040] In one embodiment of the present invention, the target vehicle surveys the entire target site and collects panoramic images of the location at multiple different locations, thereby acquiring panoramic images of the entire target area. Then, the extrinsic parameters of each camera at the center of the vehicle body are obtained through calibration, and the intrinsic parameters of each camera are used to correct image distortion, thereby obtaining a distortion-free image. Then, the image is projected onto the ground through affine transformation, and the image projected onto the ground is stitched together with the extrinsic parameters of each camera to obtain a panoramic stitched image of the target area.

[0041] Step S230: Perform semantic segmentation on the surround view stitched image of each target vehicle to obtain a vehicle segmentation image.

[0042] Step S240: The vehicle segmentation image is converted into a single-channel image according to the semantic category obtained by semantic segmentation, and then rasterized to obtain vehicle raster images of different semantic categories.

[0043] Semantic segmentation is performed on the surround-view stitched images of each target vehicle to obtain vehicle segmented images. These segmented images are then converted into single-channel images according to the semantic categories obtained from the semantic segmentation and rasterized to obtain vehicle raster images with different semantic categories. The process includes: acquiring the target vehicle surround-view stitched images, which include multiple semantic features; inputting the target vehicle surround-view stitched images into a semantic recognition model, and segmenting the target vehicle surround-view stitched images into multiple vehicle segmented images based on the semantic features; converting each of the multiple vehicle segmented images into a single-channel image according to the semantic feature category to obtain multiple target vehicle single-channel images with the same number of semantic features; and downsampling and rasterizing the multiple target vehicle single-channel images to obtain raster images with different semantic categories.

[0044] In one embodiment of the present invention, the surround-view stitched image is transmitted to a semantic recognition model such as U-net to perform semantic segmentation on the ground representation in the surround-view scene. Based on feature information such as lane lines, parking space lines, speed bumps, zebra crossings, and center lines, the corresponding pixels of the surround-view image are classified and labeled. During the segmentation process, the algorithm filters out the ground background to obtain a segmented image containing only valid semantics. The semantics are then converted into single-channel images according to categories, and morphological processing is performed on each image to further filter out invalid segmentation pixels. Afterward, the images of each channel are downsampled and then rasterized to obtain raster images with different semantics.

[0045] Step S250: Based on the pose transformation relationship between the two vehicle acquisition positions, align the vehicle raster images of different semantic categories at the two vehicle acquisition positions and stitch them together until all vehicle raster images are stitched together to obtain the target vehicle local semantic raster image.

[0046] In one embodiment of the present invention, an odometer is constructed by combining the vehicle's own wheel speedometer and inertial measurement unit (IMU) to obtain the pose transformation relationship between two observations. This relationship is then used to align and stitch the two consecutive raster images, resulting in a local semantic raster map. During the mapping process, as the vehicle drives through the mapping area, the front end automatically acquires semantic segmentation images, then rasterizes them by channel, and sends the raster images to the back end. The back end then aligns and stitches the raster images frame by frame to obtain a local semantic raster map of the target vehicle.

[0047] Step S260: Extract multiple semantic features of the target vehicle from the local semantic raster image of the target vehicle to obtain the semantic feature sequence of the target vehicle.

[0048] In one embodiment of the present invention, the pixels of the local semantic raster image of the target vehicle are semantically decomposed into channels to obtain an image channel number consistent with the number of semantic elements. Image moments (HU moments) are extracted sequentially from each channel, and then the HU moments of each semantic element are filled into an array according to the channel order to obtain the feature vector of the local semantic image.

[0049] Step S270: Based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map, similarity matching is performed to determine multiple candidate road segments. Each candidate road segment is a road segment in each preset global map.

[0050] Similarity matching is performed based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map to determine multiple candidate road segments. The preset global map includes multiple semantic features, multiple trajectory points, and lane lines of the target vehicle. The multiple trajectory points are multiple fixed points on the driving trajectory of the target vehicle. Based on the lane lines of the target vehicle and the multiple trajectory points, the preset global map is divided into multiple local maps. Based on the semantic features of the preset global map, each local map is divided into multiple local semantic maps, and multiple local single-channel images of the preset global map are obtained based on the multiple local semantic maps. The multiple local single-channel images of the global map are rasterized to obtain multiple local semantic raster images of the preset global map.

[0051] In one embodiment of the present invention, based on actual needs, the preset screenshot frame for map blocks is a rectangle with a length of 2 meters and a width of 3 meters. During the screenshot process, orientation correction is used to ensure that the overlap between two semantic blocks is minimized. When a target vehicle passes through a straight section of road, based on the target vehicle's driving trajectory, a trajectory point is set at intervals of 0.5 meters, including but not limited to points a, b, and c. The lane lines of the road section are obtained based on the vehicle's surround view image. Combining the pixel points of the lane lines, the tangent of the local lane lines is obtained. Then, the direction of the capture frame is set along the direction perpendicular to the tangent, with the midpoint connecting the two lane lines as the center of the capture frame. The orientation of the capture frame is corrected so that the capture frame is perpendicular to the two lane lines. The area is captured according to the preset rectangular capture frame to obtain local semantic blocks.

[0052] The process of determining multiple candidate road segments by performing similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map also includes: extracting the image distance of multiple single-channel images, and determining the feature vector of the local semantic raster of the preset global map based on the image distance and channel order; obtaining the semantic raster of the target vehicle's driving road based on the target vehicle's driving trajectory, and extracting the feature vector of the semantic raster of the target vehicle's driving road; matching the feature vector of the local semantic raster of the preset global map with the feature vector of the semantic raster of the target vehicle's driving road item by item based on different semantics to obtain the image distance value; when the image distance value is less than the preset threshold, it is determined that the match is successful, and multiple first candidate road segments are obtained.

[0053] In one embodiment of the present invention, feature vectors of the semantic raster image of the target vehicle's driving road are extracted; the feature vectors of the local semantic image and the feature vectors of the semantic raster image of the target vehicle's driving road are matched one by one based on different semantics to obtain multiple image distance values. The moment values ​​of the feature vectors of the local semantic image and the feature vectors of the semantic raster image of the target vehicle's driving road at points A, B, and C are H1, H2, and H3, respectively. It is verified that H1 is greater than the preset moment value K, so local A does not meet the requirements. H2 and H3 are both less than K, so the matching is successful. Points B and C are two first candidate road segments.

[0054] Step S280: The point cloud data of the segmented image based on the location of the target vehicle is matched with the candidate point cloud data of each candidate road segment to locate the target vehicle.

[0055] During vehicle movement, based on different target vehicle acquisition locations, a surrounding view mosaic image corresponding to that location is obtained. This surrounding view mosaic image undergoes semantic segmentation, rasterization, and single-channel transformation to obtain corresponding single-channel semantic raster images. Furthermore, based on these multiple semantic raster images and their semantic feature order, a set of corresponding semantic feature vectors, i.e., point cloud data, is obtained. The point cloud data of a local map of the target vehicle's location is matched with the point cloud data of the global map to obtain the point in the global map most similar to the target vehicle's location; this point is the location of the target vehicle.

[0056] The method of matching point cloud data of segmented images based on the target vehicle's location with candidate point cloud data of each candidate road segment to locate the target vehicle includes: predicting a preliminary position of the target vehicle at the moment of its current downward movement based on the target vehicle's driving trajectory; determining the feature vector of the semantic raster map of the preliminary position based on the preliminary position, and matching the feature vector of the semantic raster map of the preliminary position with the feature vector of the local semantic raster map of the complete road map one by one based on different semantics to obtain multiple second candidate road segments; comparing the relative relationship between the target vehicle's current position and the preliminary position, as well as the relative positional relationship between the multiple first candidate road segments and the multiple second candidate road segments; if the relative positional relationship between any first candidate road segment and any second candidate road segment is the same as the relative relationship between the target vehicle's current position and the preliminary position, then the road segment formed by the first road segment and the second road segment is determined to be the road segment where the target vehicle is located.

[0057] In one embodiment of the present invention, as described above, points B and C are determined to be two first candidate road segments, and the target vehicle is located at point M. Based on the target vehicle's driving trajectory, the next preliminary position of the target vehicle is point N, and point N is south of point M and adjacent to point M. Based on the feature vector of the semantic raster map of point N, second candidate road segments D and E are obtained, wherein point E is south of point B and adjacent to point B. Then, the road segment formed by points B and E is determined to be the road segment where the target vehicle is located.

[0058] It should be understood that points A, B, C, D, E, M, and N are not actual points, but rather local map tiles within the global map.

[0059] After determining the road segment where the target vehicle is located, the process also includes: acquiring lane line information on both sides of the road segment where the target vehicle is located, and determining the target vehicle and the lane line information on both sides of the road segment where the target vehicle is located based on the lane line information on both sides of the road segment where the target vehicle is located, and determining the lateral coordinate relationship between the target vehicle body and the lane; fusing the lateral coordinate relationship between the target vehicle body and the lane with the road segment information where the target vehicle is located to obtain the target vehicle positioning fusion information; and performing rasterization processing on the target vehicle positioning fusion information to obtain the target vehicle's location.

[0060] In one embodiment of the present invention, the target vehicle is known to be located on the target road segment formed by points B and E. The driving lanes on both sides of the target road segment are obtained. Facing the direction of travel of the target vehicle, with the left lane line as the origin and the direction along the right lane line as the positive X-axis, the coordinate position of the target vehicle is obtained. The coordinate information is then fused with the target road segment information, and the fused information is rasterized to obtain the accurate location of the target vehicle.

[0061] Figure 3 This is a schematic diagram illustrating a ground-based visual semantic mapping method, which is an exemplary embodiment of this application.

[0062] Before performing similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map, the process includes: acquiring multiple surround view images of the target area from multiple location acquisition points in multiple road segments of the target area, where the target area includes the driving path; stitching the multiple surround view images of the target area based on the location acquisition points to obtain a surround view stitched image of the target area from multiple location acquisition points; performing semantic segmentation on each surround view stitched image of the target area to obtain a segmented image of the area; converting the segmented image into a single-channel image according to the semantic category obtained from the semantic segmentation and performing rasterization processing to obtain a raster image of the area with different semantic categories; aligning and stitching the raster images of the area with different semantic categories from the two location acquisition points based on the pose transformation relationship between the two location acquisition points, until all the raster images of the area are stitched together to obtain a local semantic raster image of the target area; and extracting multiple semantic features of the target area from the local semantic raster image of the target area to obtain the global semantic feature sequence of the preset global map.

[0063] In one embodiment of the present invention, assuming a parking lot is taken as the target area, a vehicle is first driven around the parking lot. During the vehicle's movement, an onboard camera captures images of the surrounding environment along the vehicle's trajectory, resulting in multiple panoramic images of the parking lot. These panoramic images are then aligned and stitched together based on the vehicle's positional changes during image capture, forming a panoramic stitched image of the parking lot. Based on the semantic features of this panoramic stitched image, semantic segmentation is performed to obtain segmented images with the same number of semantic features. The resulting segmented images are then converted into single-channel images according to their corresponding semantic features, and these single-channel images are rasterized to obtain multiple semantic raster images. Subsequently, based on the vehicle's IMU and wheel speed sensor, the vehicle's pose changes during the two image captures are obtained. These pose changes are then aligned and stitched together to obtain a global map of the parking lot and its global semantic raster image.

[0064] This invention uses a low-cost surround-view camera to acquire surround-view images of the vehicle's trajectory, thereby obtaining semantic information about the ground around the vehicle. The robustness of the images is enhanced through single-channel conversion and rasterization. Furthermore, sensors such as IMUs and wheel speed sensors are used to assess changes in the vehicle's pose. Based on these changes in pose between two consecutive data acquisitions, the resulting semantic ground raster images are stitched together to create a global map of the vehicle's target path. Using semantic information as localization input effectively avoids the instability of visual feature points. Moreover, after the initial global map is generated, it can be saved to the vehicle's local storage, ensuring the long-term effectiveness of the map.

[0065] In one embodiment of the present invention, a surround view image is first acquired using a camera. The acquired surround view image is then stitched together and transmitted to a deep learning model, such as U-net, for semantic segmentation of the ground representation in the surround view scene. Subsequently, the images from each channel are downsampled and rasterized to obtain raster images with different semantic meanings. Then, an odometry system is constructed using an IMU and wheel speed sensor to obtain the pose transformation relationship between two observations. This odometry relationship is used to align and stitch the two consecutive raster images, thus completing the road map construction. The road map is then optimized to obtain a global map of the target area and saved locally. After map construction, based on the actual positional relationships of the target area and the driving trajectory of the target vehicle, the global map is divided into multiple small blocks. Raster images of each map block are obtained from the global map's raster image, forming raster map strips. Feature vectors of each raster map strip are then extracted and converted into descriptors. Finally, the raster map and descriptor sequences are saved locally on the vehicle.

[0066] Figure 4 This is a schematic diagram illustrating a ground-based visual semantic localization method as an exemplary embodiment of this application.

[0067] This invention obtains a coarse location of the vehicle in the map by comparing the semantic information around the vehicle with the semantic information of the global map. This coarse location is then used as a priori condition to narrow down the search area for grid matching, thereby quickly locating the vehicle's position in the global map.

[0068] In one embodiment of the present invention, during the driving process of the target vehicle, a surround view of the target vehicle is acquired, and based on... Figure 3 Using the same method, the surround view images acquired from the target vehicle are aligned and stitched together. This stitched surround view image is then semantically segmented and rasterized to obtain a rasterized version of the target vehicle's surround view image. Since the target vehicle's surround view image is a part of the global map, it is called a local map; therefore, the rasterized version of the target vehicle's surround view image is also called a rasterized local map. Then, based on the target vehicle's driving trajectory and the map's physical location information, this rasterized local map is physically divided into blocks to obtain rasterized local map slices, i.e., local raster map slices. Their characteristic information is extracted to obtain feature vectors, that is, the semantic raster map slices are converted into feature description sequences. Based on the obtained semantic raster map slices, this feature vector information is combined with... Figure 3 The global map and its feature vector obtained in the previous steps are compared to search for candidate points in the global map for the road segment where the target vehicle is located.

[0069] In one embodiment of the present invention, the raster image feature information of the local map of the target vehicle is compared with the raster image feature information of the global map using the Hidden Markov Model (HMM) long sequence similarity matching method to obtain candidate road segments of the target vehicle's location in the global map, and the coordinate position of the candidate road segment is calculated. The local raster image of the candidate road segment in the global map is loaded from the vehicle's local location, and the local raster image feature information of the candidate road segment is compared with the raster image feature information of the local map of the target vehicle to obtain the positioning information of the target vehicle in the global map, and the optimal matching result is selected from them.

[0070] Figure 5 This is a schematic diagram illustrating feature calculation based on semantic map tiles, as shown in an exemplary embodiment of this application.

[0071] This invention divides the global map into multiple local map tiles by truncating map segments. Then, it compares the local map of the target vehicle's location with the local map tiles of the global map to obtain the target vehicle's location. This method first performs a coarse selection of the target vehicle's location using feature vectors, and uses this location result as a priori to help narrow down the grid matching search area. Through a coarse-to-fine search and matching process, the vehicle's location on the road can be quickly obtained.

[0072] like Figure 5 As shown, in one embodiment of the present invention, when a target vehicle travels through a long straight lane, trajectory points are set along the vehicle's direction of travel to obtain the lane lines of that road segment. First, the tangent of the local lane line is obtained based on the pixels of the lane line; then, a capture frame is set along the direction perpendicular to the tangent of the lane line, with the midpoint of the line connecting the left and right lane lines as the midpoint of the capture frame, and the global map is locally divided. Assuming a trajectory point is set every 0.5 meters, and the area of ​​the capture frame is 2 square meters, that is, along the target vehicle's travel trajectory, a trajectory point is set every 0.5 meters. Based on the lane lines on both sides of the trajectory point, the tangent of the local lane line is obtained. The midpoint of the line connecting the left and right lane lines is used as the center of the capture frame, and the direction perpendicular to the local lane line is used as the long side direction of the capture frame. A 2-square-meter local map is captured between the two trajectory points as a map block. It is worth noting that during the map block capture process, the orientation of the capture frame needs to be continuously corrected so that the capture frame is perpendicular to the tangents of the lane lines on both sides, thereby obtaining multiple local map blocks with equal areas and minimal intersection areas.

[0073] Figure 6 This is a schematic diagram illustrating a slice-based semantic feature sequence, as shown in an exemplary embodiment of this application.

[0074] This invention uses semantic feature sequences as the judgment condition for localization, which enables more feature information to be utilized during localization, thereby improving its success rate and robustness.

[0075] This invention divides a global map into multiple local maps by map segmentation, and performs image segmentation, single-channel conversion, and rasterization on each local map in the same way to obtain the feature information of each map block. The feature information includes, but is not limited to, lane lines, parking space lines, speed bumps, zebra crossings, and center lines, forming a semantic feature set with road segment IDs.

[0076] In one embodiment of the present invention, semantic feature 1 is assumed to be lane line information, semantic feature 2 to be parking space line information, semantic feature 3 to be speed bump information, and semantic feature 4 to be zebra crossing information. The global map is divided into multiple map blocks, including blocks A, B, and C, using a map tiling method, and their semantic feature information is stored separately. When a local map block A is matched to determine if it is a candidate road segment for the target vehicle, the raster semantic information of the target vehicle's surround-view mosaic image is compared one by one with the raster semantic information of the map block A. First, the lane line information of the target vehicle surround view mosaic raster semantic information is compared with the lane line information of the A map block raster semantic information. Then, the location line information of the target vehicle surround view mosaic raster semantic information is compared with the location line information of the A map block raster semantic information. This process is repeated, comparing each semantic feature of the target vehicle surround view mosaic raster semantic information and the A map block raster semantic information one by one to obtain the HU value of the A map block, thereby determining whether the A map block is the location of the target vehicle. This process is repeated for the B map block and the C map block until the location of the target vehicle is determined.

[0077] Figure 7 This is a block diagram illustrating a vehicle positioning device according to an exemplary embodiment of this application. The device can be applied to… Figure 1 The implementation environment shown is specifically configured in the smart terminal 102. This device can also be applied to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which the device is applicable.

[0078] like Figure 7 As shown, the exemplary vehicle positioning device includes: an image acquisition module 710, a first image stitching module 720, an image segmentation module 730, an image conversion module 740, a second image stitching module 750, a feature extraction module 760, a feature matching module 770, and a vehicle positioning module 780.

[0079] The system includes: an image acquisition module 710 for acquiring multiple surround view images of the target vehicle at multiple vehicle acquisition locations along the target vehicle's driving path; a first image stitching module 720 for stitching the multiple surround view images of the target vehicle based on the vehicle acquisition locations to obtain a stitched surround view image of the target vehicle at multiple vehicle acquisition locations; an image segmentation module 730 for performing semantic segmentation on each stitched surround view image of the target vehicle to obtain a segmented vehicle image; an image conversion module 740 for converting the segmented image into a single-channel image according to the semantic category obtained from the semantic segmentation, and performing rasterization processing to obtain vehicle raster images of different semantic categories; and a second image stitching module 750 for stitching the images based on the pose transformation relationship between the front and rear vehicle acquisition locations. Vehicle raster images of different semantic categories at two vehicle acquisition locations are aligned and stitched together until all vehicle raster images are stitched together to obtain a local semantic raster image of the target vehicle; feature extraction module 760 is used to extract multiple target vehicle semantic features from the local semantic raster image of the target vehicle to obtain a target vehicle semantic feature sequence; feature matching module 770 is used to perform similarity matching based on the target vehicle semantic feature sequence and the global semantic feature sequence of a preset global map to determine multiple candidate road segments, where each candidate road segment is one of the road segments in each preset global map; vehicle localization module 780 is used to match the point cloud data of the segmented image at the target location with the candidate point cloud data of each candidate road segment to locate the target vehicle.

[0080] It should be noted that the vehicle positioning device and the vehicle positioning method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the vehicle positioning device provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0081] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the vehicle positioning method provided in the above embodiments.

[0082] Figure 8 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown. It should be noted that... Figure 8 The computer system 800 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of the embodiments of this application.

[0083] like Figure 8As shown, the computer system 800 includes a Central Processing Unit (CPU) 801, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on a program stored in Read-Only Memory (ROM) 802 or a program loaded from storage portion 808 into Random Access Memory (RAM) 803. The RAM 803 also stores various programs and data required for system operation. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An Input / Output (I / O) interface 805 is also connected to the bus 804.

[0084] The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 810 as needed so that computer programs read from it can be installed into storage section 808 as needed.

[0085] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit (CPU) 801, it performs various functions defined in the system of this application.

[0086] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0087] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0088] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0089] Another aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a computer's processor, causes the computer to perform the vehicle positioning method as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not incorporated into the electronic device.

[0090] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the vehicle positioning method provided in the various embodiments described above.

[0091] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A vehicle positioning method, characterized in that, The method includes: Acquire multiple surround view images of the target vehicle at multiple vehicle collection locations along the target vehicle's driving path; Multiple surround view images of the target vehicle are stitched together based on the vehicle acquisition location to obtain a stitched surround view image of the target vehicle at multiple vehicle acquisition locations. Semantic segmentation is performed on the surround view stitched images of each target vehicle to obtain vehicle segmentation images; The vehicle segmentation image is converted into a single-channel image according to the semantic category obtained by semantic segmentation, and then rasterized to obtain vehicle raster images of different semantic categories. Based on the pose transformation relationship between the two vehicle acquisition positions, the vehicle grid images of different semantic categories at the two vehicle acquisition positions are aligned and stitched together until all the vehicle grid images are stitched together to obtain the target vehicle semantic grid image of the target vehicle. Extract multiple target vehicle semantic features from the target vehicle semantic raster image to obtain a target vehicle semantic feature sequence; Based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map, similarity matching is performed to determine multiple candidate road segments, wherein the candidate road segment is one of the road segments in the preset global map; The point cloud data of the segmented image based on the location of the target vehicle is matched with the candidate point cloud data of each candidate road segment to locate the target vehicle.

2. The vehicle positioning method as described in claim 1, characterized in that, Before performing similarity matching based on the target vehicle semantic feature sequence and the global semantic feature sequence of the preset global map, the method further includes: Acquire multiple panoramic images of the target area from multiple locations within multiple road segments of the target area, wherein the target area includes the driving path; Multiple surround view images of the target area are stitched together based on the area acquisition location to obtain a target area surround view stitched image with multiple area acquisition locations; Semantic segmentation is performed on the surround view stitched images of each target region to obtain region segmentation images; The segmented image is converted into a single-channel image according to the semantic category obtained by semantic segmentation, and then rasterized to obtain raster images of regions with different semantic categories; Based on the pose transformation relationship between the two regions, the raster maps of different semantic categories at the two regions are aligned and stitched together until all the raster maps are stitched together to obtain the target region local semantic raster map of the target region. Extract multiple target region semantic features from the local semantic raster map of the target region to obtain a global semantic feature sequence of the preset global map.

3. The vehicle positioning method as described in claim 1, characterized in that, Multiple surround view images of the target vehicle are stitched together based on the vehicle acquisition location to obtain a stitched surround view image of the target vehicle at the vehicle acquisition location, including: Multiple surround view images of the target vehicle are captured using a camera, and image distortion correction is performed on the surround view images of the target vehicle based on the extrinsic and intrinsic parameters of the camera to obtain a distortion-free surround view image of the target vehicle. The multiple surround view images of the target vehicle include at least one of the front image, rear image, left image, and right image of the target vehicle. The distortion-free target vehicle surround view image is projected onto the ground through an affine transformation to obtain a road top view image; The road top view image is stitched together based on the extrinsic and intrinsic parameters of the camera to obtain a surround view stitched image of the target vehicle.

4. The vehicle positioning method as described in claim 1, characterized in that, Semantic segmentation is performed on the surround-view stitched images of each target vehicle to obtain vehicle segmentation images. These vehicle segmentation images are then converted into single-channel images according to the semantic categories obtained from the semantic segmentation, and then rasterized to obtain vehicle raster images with different semantic categories, including: Obtain a surround view stitched image of the target vehicle, wherein the surround view stitched image of the target vehicle includes multiple semantic features; The target vehicle surround view stitched image is input into a semantic recognition model, and the target vehicle surround view stitched image is segmented into multiple vehicle segmentation images according to the semantic features; The multiple vehicle segmentation images are converted into single-channel images according to the semantic feature categories to obtain multiple target vehicle single-channel images with the same number of semantic features. The single-channel images of the multiple target vehicles are downsampled and rasterized to obtain raster images with different semantics.

5. The vehicle positioning method as described in claim 1, characterized in that, Based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map, similarity matching is performed to determine multiple candidate road segments, including: The preset global map includes multiple semantic features, multiple trajectory points, and target vehicle lane lines, wherein the multiple trajectory points are multiple fixed points on the target vehicle's driving trajectory; Based on the target vehicle lane lines and the multiple trajectory points, the preset global map is divided into multiple preset global map local maps; Based on the semantic features of the preset global map, the local map of the preset global map is segmented into multiple local semantic maps of the preset global map, and multiple local single-channel images of the preset global map are obtained based on the multiple local semantic maps of the preset global map. The multiple preset global map local single-channel images are rasterized to obtain multiple preset global map local semantic raster images.

6. The vehicle positioning method as described in claim 1, characterized in that, Based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map, similarity matching is performed to determine multiple candidate road segments, which also includes: Extract the image distances from the multiple single-channel images, and determine the feature vectors of the preset global map local semantic raster based on the image distances and channel order; Based on the target vehicle's driving trajectory, a semantic raster map of the target vehicle's driving road is obtained, and the feature vector of the semantic raster map of the target vehicle's driving road is extracted. The feature vectors of the preset global map local semantic raster and the feature vectors of the target vehicle driving road semantic raster are matched item by item based on different semantics to obtain the image distance value; If the image distance value is less than a preset threshold, it is determined that the match is successful, and multiple first candidate road segments are obtained.

7. The vehicle positioning method as described in claim 5, characterized in that, Matching point cloud data from segmented images based on the target vehicle's location with candidate point cloud data from each of the candidate road segments to locate the target vehicle includes: Based on the target vehicle's driving trajectory, predict the target vehicle's preparatory position at the moment of its current downward movement; Based on the proposed location, the feature vector of the proposed location semantic raster is determined, and the feature vector of the proposed location semantic raster is matched one by one with the feature vector of the local semantic raster of the complete road map based on different semantics to obtain multiple second candidate road segments. The relative positions of the target vehicle's current position and its intended position are compared with the relative positions of multiple first candidate road segments and multiple second candidate road segments. If the relative position of any first candidate road segment and any second candidate road segment is the same as the relative position of the target vehicle's current position and its intended position, then the road segment formed by the first candidate road segment and the second candidate road segment is determined to be the road segment where the target vehicle is located.

8. The vehicle positioning method according to any one of claims 1-7, characterized in that, After determining the road segment where the target vehicle is located, the following is also included: The lane line information on both sides of the road segment where the target vehicle is located is obtained, and the lane line information on both sides of the road segment where the target vehicle is located is determined based on the lane line information on both sides of the road segment where the target vehicle is located, and the lateral coordinate relationship between the target vehicle body and the lane is determined. The lateral coordinate relationship between the target vehicle body and the lane is fused with the road segment information where the target vehicle is located to obtain the target vehicle positioning fusion information; The target vehicle's location fusion information is rasterized to obtain the target vehicle's location.

9. A vehicle positioning device, characterized in that, include: The image acquisition module is used to acquire multiple surround view images of the target vehicle at multiple vehicle collection locations along the target vehicle's driving path; The first image stitching module is used to stitch together multiple target vehicle surround view images based on the vehicle acquisition location to obtain a target vehicle surround view stitched image with multiple vehicle acquisition locations. The image segmentation module is used to perform semantic segmentation on the surround view stitched images of each of the target vehicles to obtain vehicle segmentation images; The image conversion module is used to convert the vehicle segmentation image into a single-channel image according to the semantic category obtained by semantic segmentation, and to perform rasterization processing to obtain vehicle raster images of different semantic categories. The second image stitching module is used to align and stitch vehicle raster images of different semantic categories at the two vehicle acquisition positions based on the pose transformation relationship between the two vehicle acquisition positions, until all the vehicle raster images are stitched together to obtain the target vehicle local semantic raster image of the target vehicle. The feature extraction module is used to extract multiple semantic features of the target vehicle from the local semantic raster image of the target vehicle to obtain a semantic feature sequence of the target vehicle. The feature matching module is used to perform similarity matching based on the semantic feature sequence of the target vehicle and the global semantic feature sequence of the preset global map to determine multiple candidate road segments, wherein the candidate road segment is a road segment in the preset global map; The vehicle localization module is used to match the point cloud data of the segmented image based on the target location with the candidate point cloud data of each candidate road segment in order to locate the target vehicle.

10. An electronic device, characterized in that, The electronic device includes: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the vehicle positioning method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the vehicle positioning method according to any one of claims 1 to 8.

Images