Map construction method, vehicle and storage medium

Abstract

The embodiment of the invention provides a map construction method, a vehicle and a storage medium, and the method comprises the steps: obtaining multi-frame point cloud data of the vehicle in response to the operation of the vehicle on a target operation path, the point cloud data being used for representing the pose state of the vehicle and the structure information of a map where the target operation path is located; multiple frames of target point cloud data are screened out from the multiple frames of point cloud data, and the pose state corresponding to the target point cloud data meets the target pose state; determining a plurality of target feature points from the plurality of feature points of the target point cloud data; and based on the plurality of target feature points corresponding to the multi-frame target point cloud data, constructing map data of a map. The technical problem of low accuracy of map construction is solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a map construction method, a vehicle, and a storage medium. Background Technology

[0002] Currently, vehicle identification on its operating path typically uses GPS positioning data for simple route matching or builds historical driving routes based on high-precision maps. However, this method suffers from low positioning accuracy and has technical problems with low map accuracy.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] This application provides a map construction method, a vehicle, and a storage medium to at least address the technical problem of low accuracy in map construction.

[0005] According to one aspect of the embodiments of this application, a map construction method is provided. The method may include: in response to a vehicle running on a target running path, acquiring multi-frame point cloud data of the vehicle, wherein the point cloud data is used to characterize the vehicle's pose state and the structural information of the map where the target running path is located; filtering out multi-frame target point cloud data from the multi-frame point cloud data, wherein the pose state corresponding to the target point cloud data satisfies the target pose state; determining multiple target feature points from multiple feature points of the target point cloud data; and constructing map data of the map based on the multiple target feature points corresponding to the multi-frame target point cloud data.

[0006] Furthermore, multiple frames of target point cloud data are selected from the multi-frame point cloud data, including: determining the vehicle's pose state based on the point cloud data; and determining the point cloud data as target point cloud data in response to the pose state satisfying the target pose state, wherein the target pose state is used to characterize the state in which the vehicle's pose change exceeds the adjustment threshold.

[0007] Furthermore, from multiple feature points in the target point cloud data, multiple target feature points are determined, including: determining the average gradient corresponding to each of the multiple feature points, and determining at least one edge feature point and at least one planar feature point among the multiple feature points, wherein the average gradient is used to characterize the intensity change characteristics of the feature point; in response to the average gradient being greater than the intensity gradient threshold, the feature point is determined as a target feature point, and the edge feature point or the planar feature point is determined as a target feature point.

[0008] Furthermore, based on multiple target feature points corresponding to multiple frames of target point cloud data, map data is constructed to obtain a map, including: transforming multiple target feature points corresponding to the target point cloud data to obtain a relative transformation matrix corresponding to the target point cloud data, wherein the relative transformation matrix is ​​used to characterize the pose change state of the target feature points; converting multiple target feature points corresponding to multiple frames of target point cloud data into multiple first feature points in the same coordinate system, wherein the first feature points correspond to the target feature points; and using the relative transformation matrix to stitch together the multiple first feature points to obtain map data.

[0009] Further, the multiple target feature points corresponding to the target point cloud data are transformed to obtain the relative change matrix corresponding to the target point cloud data, including: determining the second target point cloud data adjacent to the first target point cloud data, wherein the multi-frame target point cloud data includes the first target point cloud data and the second target point cloud data; in response to the average gradient corresponding to the first target feature point in the first target point cloud data being greater than the intensity gradient threshold, or the first target feature point being an edge feature point, determining the third target feature point adjacent to the first target feature point and the fourth target feature point next to the first target feature point among the multiple second target feature points corresponding to the second target point cloud data; constructing a line feature based on the third target feature point and the fourth target feature point; obtaining the first distance function between the line feature and the first target feature point; and constructing the relative change matrix based on the first distance function.

[0010] Further, based on the first distance function, a relative change matrix is ​​constructed, including: in response to the first target feature point being a planar feature point, among multiple second target feature points corresponding to the second target point cloud data, determining a fifth target feature point adjacent to the first target feature point, a sixth target feature point adjacent to the first target feature point at the same scan line number, and a seventh target feature point adjacent to the scan line adjacent to the fifth target feature point; constructing a surface feature based on the fifth target feature point, the sixth target feature point, and the seventh target feature point; obtaining a second distance function between the surface feature and the second target feature point; and constructing a relative change matrix based on the first distance function and the second distance function.

[0011] Furthermore, based on the first distance function and the second distance function, a relative change matrix is ​​constructed, including: weighting and fusing the first distance function and the second distance function to obtain the target distance function; and performing nonlinear processing on the target distance function to obtain the relative change matrix.

[0012] Furthermore, the method may also include: in response to the vehicle running on the target running path, acquiring the vehicle's target positioning information and the vehicle's running data, wherein the running data is used to characterize the vehicle's running state; constructing a graph model using multi-frame point cloud data, target positioning information, and running data, wherein the graph model includes multiple state nodes, the state nodes are used to characterize the vehicle's pose state at different times, and the connecting edges between state nodes are used to characterize the constraint relationships between state nodes; and using the graph model to determine the vehicle's position information in the map data.

[0013] According to another aspect of the embodiments of this application, a processor load rate determination map construction apparatus is also provided. The apparatus may include: an acquisition unit, configured to acquire multi-frame point cloud data of a vehicle in response to a vehicle running on a target running path, wherein the point cloud data is used to characterize the vehicle's pose state and the structural information of the map where the target running path is located; a filtering unit, configured to filter out multiple frames of target point cloud data from the multi-frame point cloud data, wherein the pose state corresponding to the target point cloud data satisfies the target pose state; a determination unit, configured to determine multiple target feature points from multiple feature points of the target point cloud data; and a construction unit, configured to construct map data of a map based on the multiple target feature points corresponding to the multi-frame target point cloud data.

[0014] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of this application when it runs.

[0015] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to run the methods of various embodiments of this application.

[0016] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that implements the methods of various embodiments of this application when run by a processor.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which is executed by a processor to implement the methods in various embodiments of this application.

[0018] According to another aspect of the embodiments of this application, a computer program is also provided, which is executed by a processor to implement the methods of the various embodiments of this application.

[0019] In this embodiment, in response to a vehicle running on a target path, multi-frame point cloud data of the vehicle is acquired. The point cloud data represents the vehicle's pose state and the structural information of the map where the target path is located. From the multi-frame point cloud data, multiple frames of target point cloud data are selected, where the pose state corresponding to the target point cloud data satisfies the target pose state. Multiple target feature points are determined from multiple feature points in the target point cloud data. Based on the multiple target feature points corresponding to the multi-frame target point cloud data, map data is constructed. That is, in this application, multi-frame point cloud data of the vehicle is acquired, and multiple frames of target point cloud data are determined. These target point cloud data can be keyframes in the multi-frame point cloud data. Based on the target feature points in the keyframes, map data can be constructed. This map data can be a high-precision map containing the vehicle's operating area, thereby improving the accuracy of map construction and solving the technical problem of low accuracy in map construction. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This is a flowchart of a map construction method according to an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of edge feature point or intensity feature point matching according to an embodiment of this application;

[0023] Figure 3 This is a schematic diagram of planar feature point matching according to an embodiment of this application;

[0024] Figure 4 This is a flowchart of a common driving route identification and optimization strategy according to an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of a map construction apparatus according to an embodiment of this application. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] According to an embodiment of this application, a method embodiment for constructing a map is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] Currently, vehicle driving route recognition typically relies on GPS positioning data to match historical routes. This method suffers from low positioning accuracy, and its effectiveness is significantly affected by environmental factors such as tunnels and trees. Furthermore, network transmission speed also impacts positioning performance. The root cause lies in the lack of semantic-level scene feature extraction during positioning.

[0030] In related technologies, Simultaneous Localization and Mapping (SLAM) algorithms are commonly used to extract semantic scene features and construct high-precision maps for scene localization. Alternatively, feature matching can be achieved by extracting planar points and edge points based on point cloud curvature features. However, these methods are hampered by the diverse nature of real-world scenes, such as areas combining open and built-up environments. When LiDAR acquires point cloud data, the point cloud distribution is uneven, and feature points are scattered, leading to suboptimal feature point extraction. This reduces the accuracy of point cloud matching, significantly impacting the positioning accuracy and mapping performance of LiDAR odometry, resulting in low map accuracy.

[0031] To address the aforementioned issues, this embodiment acquires multi-frame point cloud data of the vehicle and determines multi-frame target point cloud data within the multi-frame point cloud data. This target point cloud data can be keyframes within the multi-frame point cloud data. Based on the target feature points in the multi-frame keyframes, map data can be constructed. This map data can be a high-precision map containing the vehicle's operating area, thereby optimizing the construction of high-precision maps for commonly used driving routes and improving vehicle positioning accuracy. This avoids problems such as low efficiency in identifying commonly used driving routes, thus achieving the technical effect of improving the accuracy of map construction and solving the technical problem of low accuracy in map construction.

[0032] This embodiment provides a method for constructing a map. Figure 1 This is a flowchart of a map construction method according to an embodiment of this application. Figure 1 As shown, the method may include the following steps.

[0033] Step S102: In response to the vehicle running on the target running path, acquire multi-frame point cloud data of the vehicle, wherein the point cloud data is used to characterize the vehicle's pose state and the structural information of the map where the target running path is located.

[0034] In the technical solution provided in step S102 of this application, the target running path can be a commonly used driving path of the vehicle, or a path with a high driving frequency. The multi-frame point cloud data can be data collected by the lidar sensor in the vehicle, and can include multiple feature points. It can be a dataset including multiple three-dimensional coordinate data, which can be used to determine the vehicle's pose state and the structural information of the map where the target running path is located. The feature points can be three-dimensional coordinate data. The pose state can be the vehicle's pose change, and can be used to characterize data such as the vehicle's movement or rotation angle.

[0035] Optionally, in autonomous driving scenarios, vehicles can be equipped with LiDAR sensors, which can capture real-time three-dimensional information about the vehicle's surrounding environment. LiDAR measures the distance between objects and the sensor by emitting laser beams and receiving reflected signals, generating a series of three-dimensional coordinate data of points; this data set is called point cloud data.

[0036] Optionally, when the vehicle is traveling on the target path, radar can be used to acquire multiple frames of point cloud data of the vehicle during its movement. Based on this multi-frame point cloud data, the vehicle's pose state during operation can be determined, and by identifying the point cloud data, structural information of the map where the target path is located can be obtained.

[0037] Optionally, to preliminarily determine whether a certain path is a frequently used driving route for a vehicle, GPS positioning information of the vehicle before driving and after parking within the past month can be collected to obtain high-frequency positioning points, which can be used as the start and end points of the user's frequently used driving route. When the start and end points of the vehicle's current driving route are consistent with the vehicle's frequently used driving route, it can be determined that the vehicle is driving on the target driving route, and at this time, multi-frame point cloud data of the vehicle can be acquired.

[0038] Alternatively, although point cloud data directly records the structural information of the environment, by matching consecutive point cloud frames (i.e., multiple frames of point cloud data), the vehicle's movement and rotation in space can be inferred. By calculating the relative change matrix between adjacent point cloud frames, the vehicle's translational and rotational displacements can be determined, thereby updating the vehicle's real-time pose information.

[0039] Optionally, point cloud data can also be used to construct and update maps of the environment in which a vehicle travels. Through feature point extraction (e.g., planar and edge features) and map registration algorithms, a high-precision map containing detailed environmental structures such as roads, obstacles, and landmarks can be gradually built. Such a map is crucial for vehicle path planning, obstacle avoidance, and positioning, especially when the vehicle needs to travel in environments with no or poor Global Navigation Satellite System (GNSS) signal.

[0040] In this embodiment, the multi-frame point cloud data acquired by the LiDAR can not only estimate the vehicle's pose (position and orientation) in real time, but also progressively build and update the environmental map of the vehicle's driving path. This Simultaneous Localization and Mapping (SLAM) capability is one of the key technologies for realizing autonomous navigation of autonomous vehicles, allowing the vehicle to dynamically understand its position and make appropriate decisions and adjustments in unknown or partially known environments.

[0041] Through the above process, the autonomous driving system can respond to the actual operation of the vehicle on the target path, update the vehicle's position and posture in real time, and continuously enrich and refine the map information of the surrounding environment, providing the necessary environmental perception and positioning foundation for the vehicle's intelligent driving.

[0042] Step S104: Select multiple frames of target point cloud data from the multiple frames of point cloud data, wherein the pose state corresponding to the target point cloud data satisfies the target pose state.

[0043] In the technical solution provided in step S104 of this application, the target point cloud data can be keyframes, and the vehicle's pose state in the target point cloud data satisfies the target pose state. This target pose state can be a pre-set state, such as the vehicle's pose change exceeding a translation threshold of 1.5 meters (m) or a rotation threshold of 15 degrees (°).

[0044] Optionally, based on continuous multi-frame point cloud data, the vehicle's pose state in each frame of point cloud data can be determined. When the vehicle's pose state meets the target pose state, the point cloud data can be identified as the target point cloud data. Through the above steps, multiple frames of target point cloud data can be determined from the multi-frame point cloud data.

[0045] Optionally, the aforementioned multi-frame target point cloud data can be a set of multiple keyframes that are consecutive on the time axis.

[0046] Step S106: Determine multiple target feature points from multiple feature points in the target point cloud data.

[0047] In the technical solution provided in step S106 of this application, the target feature points may include, but are not limited to, edge feature points, planar feature points, and intensity feature points. Edge feature points may refer to locations in point cloud data situated at object boundaries or surface discontinuities. Planar feature points may refer to points located on large, flat surfaces. Intensity feature points may refer to points with significant reflection intensity, or points with unique intensity values ​​at the boundaries of different materials or environments.

[0048] Optionally, multiple target feature points can be determined from a pool of feature points. These target feature points may include, but are not limited to, edge feature points, planar feature points, and intensity feature points. Edge feature points and planar feature points can identify object boundaries and large flat surfaces, respectively, based on the geometric distribution of the point cloud data. Intensity feature points can utilize variations in reflection intensity from the point cloud data to provide additional positioning information at material boundaries or complex structures, enhancing the map's information content and positioning accuracy.

[0049] Optionally, target feature points are extracted from the acquired target point cloud data. This embodiment innovatively introduces intensity feature points. The extraction of intensity feature points is based on the change in point cloud intensity gradient. Nearest neighbor points are searched using a K-Dimensional Tree (KD-tree), and the average intensity gradient is calculated. When the gradient exceeds a set threshold, the point is determined to be an intensity feature point. Simultaneously, points that overlap with edge feature points need to be excluded based on a curvature threshold to ensure the independence of intensity feature points.

[0050] Optionally, edge feature points and planar feature points can be determined by analyzing the geometric properties of the point cloud data returned by the LiDAR. The extraction of edge feature points and planar feature points is crucial for the SLAM algorithm in the process of environmental perception and localization.

[0051] For example, edge feature points can be extracted through the following steps: For each feature point in the target point cloud data, the nearest neighbors of the point can be quickly found by constructing a KD-tree or other data structures to calculate the distribution of points within the neighborhood of that feature point. Using the points' neighbors, the local curvature of the point is calculated. Curvature can be calculated using various mathematical methods, such as fitting a local plane or cylinder. A curvature threshold is set; points with curvature higher than this threshold are considered edge feature points. These points are typically located at inflection points on object surfaces or as straight lines in the environment.

[0052] For another example, planar feature points can refer to points that lie on a relatively flat surface, such as the ground. These points have low local curvature and can be used to characterize large-scale environmental structures. Planar feature point extraction can be performed by calculating the neighborhood of each point in the point cloud. Using the neighborhood points, a set of points is fitted to a planar model using least squares or other fitting methods. The distance residual between each point and its fitted plane is calculated. Points with small distance residuals are considered to lie on the plane. A distance threshold is set; points with distances less than this threshold are considered planar feature points. Planar feature points can be used to construct stable environmental references, especially on large-scale or less textured planes.

[0053] It should be noted that the determination of the edge feature points and planar feature points mentioned above is only for illustrative purposes and no specific restrictions are imposed here.

[0054] Optionally, the point cloud data generated by LiDAR when scanning the environment can contain a large number of feature points. However, not all feature points are equally important. Target feature points with unique geometric characteristics can be selected from multiple feature points. These target feature points can be edge points, planar points, or intensity feature points in the point cloud data. They can be used to characterize key structures in the map, such as walls, ground, and object boundaries, thereby aiding in subsequent mapping and localization processes.

[0055] Step S108: Based on multiple target feature points corresponding to multiple frames of target point cloud data, construct map data to obtain the map.

[0056] In the technical solution provided in step S108 of this application, multiple target feature points are determined in each target point cloud data. Based on the multiple target feature points corresponding to the multi-target point cloud data, map data containing the target's running path can be constructed. The aforementioned map data can be a high-precision map.

[0057] Optionally, in response to the vehicle running on the target path, multi-frame point cloud data of the vehicle is acquired. This point cloud data can be collected by sensors such as LiDAR and can be used to characterize the vehicle's pose state (e.g., vehicle position, orientation, etc.) and the structural information of the map where the target path is located (e.g., the position and shape of roads, buildings, obstacles, etc.). From the multi-frame point cloud data, multiple frames of target point cloud data are selected. The selection criteria are that the pose state corresponding to these point cloud data must meet the target pose state. Selection removes point cloud data that does not meet the requirements, improving the effectiveness of subsequent processing and the accuracy of map construction. Furthermore, multiple target feature points can be determined from the multiple feature points of the target point cloud data. Based on the multiple target feature points corresponding to the multi-frame target point cloud data, map data can be constructed. This step integrates and processes the previously selected and determined target feature points, using point cloud registration, fusion, and other algorithms to generate map data that accurately describes the map where the target path is located.

[0058] Through steps S102 to S108, in response to the vehicle running on the target running path, multi-frame point cloud data of the vehicle is acquired. The point cloud data represents the vehicle's pose state and the structural information of the map where the target running path is located. From the multi-frame point cloud data, multiple frames of target point cloud data are selected, where the pose state corresponding to the target point cloud data satisfies the target pose state. Multiple target feature points are determined from multiple feature points in the target point cloud data. Based on the multiple target feature points corresponding to the multi-frame target point cloud data, map data is constructed. That is, in this application, multi-frame point cloud data of the vehicle is acquired, and multiple frames of target point cloud data are determined from the multi-frame point cloud data. These target point cloud data can be keyframes in the multi-frame point cloud data. Based on the target feature points in the multiple keyframes, map data can be constructed. This map data can be a high-precision map containing the vehicle's running area, thereby achieving the technical effect of improving the accuracy of map construction and solving the technical problem of low accuracy in map construction.

[0059] The above-mentioned method of this application will be further described below.

[0060] As an optional implementation, step S104 involves filtering out multiple frames of target point cloud data from the multiple frames of point cloud data, including: determining the vehicle's pose state based on the point cloud data; and determining the point cloud data as target point cloud data in response to the pose state satisfying the target pose state, wherein the target pose state is used to characterize the state in which the vehicle's pose change exceeds an adjustment threshold.

[0061] In this embodiment, the aforementioned adjustment threshold can be a pre-set translation threshold or rotation threshold, such as a translation threshold of 1.5m or a rotation threshold of 15°. It should be noted that this is merely an illustrative example, and there are no specific limitations on the size or type of the adjustment threshold.

[0062] Optionally, the target pose state can be preset to a translation threshold of 1.5m or a rotation threshold of 15°. After acquiring multiple frames of point cloud data, the pose state of the vehicle in each frame can be determined. If the pose state satisfies the target pose state, that is, if the vehicle's pose change exceeds the translation threshold of 1.5m or the rotation threshold of 15°, the point cloud data of the current frame can be determined as the target point cloud data. Through the above steps, multiple frames of target point cloud data can be determined from multiple frames of point cloud data.

[0063] For example, this embodiment proposes a keyframe pose estimation method for LiDAR odometry based on a sliding window. This method uses a fixed-length sliding window to store LiDAR frames (i.e., multiple frames of point cloud data). When the vehicle's pose change exceeds a translation threshold of 1.5m or a rotation threshold of 15°, the point cloud data of the current frame (also known as the current frame) can be determined as a keyframe (i.e., target point cloud data). The above method can obtain a set of 25 consecutive keyframes on the time axis, and pose estimation can be performed based on multiple keyframes.

[0064] As an optional implementation, step S106 involves determining multiple target feature points from multiple feature points in the target point cloud data, including: determining the average gradient corresponding to each of the multiple feature points, and determining at least one edge feature point and at least one planar feature point among the multiple feature points, wherein the average gradient is used to characterize the intensity change characteristics of the feature points; in response to the average gradient being greater than the intensity gradient threshold, the feature point is determined as a target feature point, and the edge feature point or the planar feature point is determined as a target feature point.

[0065] In this embodiment, the average gradient corresponding to each feature point can be determined. If the average gradient is greater than the intensity gradient threshold, the feature point can be identified as a target feature point. Simultaneously, edge feature points and / or planar feature points among multiple feature points can be identified, and these edge feature points and / or planar feature points can be identified as target feature points.

[0066] Optionally, when the user travels the target path again, the SLAM algorithm can be used to map the path scene to obtain a high-precision map of the scene. This embodiment introduces intensity feature points to improve point cloud registration accuracy, based on traditional edge feature point and planar feature point extraction.

[0067] Optionally, the concept of intensity gradient is introduced for the extraction of intensity feature points. Each feature point in the target point cloud data to be extracted can be traversed, and a KD-tree nearest neighbor search is used with a search radius of 10 nearest neighbor points (excluding the current point itself). The searched neighbor point indices are traversed, and the absolute difference between the intensity of the current point (i.e., the traversed feature point) and the intensity of its neighbors is calculated. These differences are accumulated in the gradient variable to obtain the average gradient corresponding to that feature point. This average gradient can be the ratio of the accumulated gradient to the number of neighbor points.

[0068] Optionally, if the average gradient corresponding to a feature point is greater than the intensity gradient threshold, then the feature point can be determined as an intensity feature point and can be identified as a target feature point. The average gradient can be calculated using the following formula:

[0069]

[0070] In the formula: It can be used to represent the intensity gradient value at point i; It can be used to represent the intensity gradient threshold; L can be a set of points on the same scan line; S can be a point i in set L and multiple points around it; j can be a point in S; I can be the depth value corresponding to the feature point; k can be the scan frame index.

[0071] Optionally, the magnitude of the intensity gradient threshold can be preset according to actual needs.

[0072] Optionally, in large-scale environments, the extracted intensity feature points can be located at the edge regions where different materials intersect, as well as at points with complex surrounding environments. Intensity feature points can be located where the surface properties of an object change, such as at the boundary between two materials with different reflectivities, or in areas where multiple structures intertwine in the scene. The curvature of the intensity feature points extracted from these areas may also meet the criteria for edge feature points. However, it is unreasonable and meaningless to replace edge feature points with intensity feature points. Therefore, in addition to satisfying the requirement that the average gradient is greater than the intensity gradient threshold, intensity feature points also need to meet the following requirements:

[0073]

[0074] Among them, intensity gradient threshold It can be 20; The curvature can be defined at a certain point. This is the curvature threshold.

[0075] In this embodiment, a lidar odometry based on curvature and point cloud intensity is designed to acquire the precise pose of the vehicle body at each moment. By extracting features from the target point cloud data of each frame, planar points, edge points, and intensity feature points are extracted to obtain multiple target feature points corresponding to that frame's point cloud data. These multiple target feature points are then matched and accumulated with a local map to construct a high-precision map. The aforementioned local map can be constructed based on multiple target feature points corresponding to previous target point cloud data of that frame.

[0076] Optionally, this embodiment incorporates point cloud intensity feature points to obtain more detailed point cloud information, addressing the issue of unclear point cloud features in large scenes. By adding intensity feature points, the matching accuracy between keyframes containing these feature points and the local map becomes more precise, enabling the construction of richer and more accurate high-precision maps and the acquisition of more accurate vehicle pose information for more accurate identification of users' frequently used driving routes.

[0077] As an optional implementation, step S108, constructing map data based on multiple target feature points corresponding to multiple frames of target point cloud data, includes: transforming the multiple target feature points corresponding to the target point cloud data to obtain a relative transformation matrix corresponding to the target point cloud data, wherein the relative transformation matrix is ​​used to characterize the pose change state of the target feature points; converting the multiple target feature points corresponding to the multiple frames of target point cloud data into multiple first feature points in the same coordinate system, wherein the first feature points correspond to the target feature points; and using the relative transformation matrix to stitch together the multiple first feature points to obtain map data.

[0078] In this embodiment, the aforementioned relative change matrix can be used to characterize the pose change state of the target point cloud data, and can be constructed based on multiple target feature points.

[0079] Optionally, by transforming multiple target feature points corresponding to the target point cloud data, a relative transformation matrix corresponding to the target point cloud data can be obtained. Multiple target feature points corresponding to multiple frames of target point cloud data can be converted into multiple first feature points in the same coordinate system. These first feature points correspond one-to-one with the target feature points, and can be points in the world coordinate system. The relative transformation matrix can then be used to stitch together these multiple first feature points to obtain map data.

[0080] Optionally, this embodiment uses a fixed-length sliding window to store laser frames. When the vehicle pose change exceeds a translation threshold of 1.5m or a rotation threshold of 15°, the current frame is determined to be a keyframe. Pose estimation is performed using a set of 25 consecutive keyframes on the time axis, establishing a set of transformation relationships between keyframes and the world coordinate system W. This transforms the target feature points in each keyframe to the W coordinate system, obtaining the first feature points of multiple target point cloud data in the W coordinate system. Multiple first feature points can be stitched together to generate a local map containing edge features, planar features, and intensity features. The stitched local map is downsampled, where edge features and intensity features can use a resolution of 0.2m, and planar features can use a resolution of 0.4m, thereby effectively eliminating redundant feature points and achieving lightweight real-time localization and mapping.

[0081] For example, multiple target feature points corresponding to multiple frames of target point cloud data can be transformed to obtain a relative transformation matrix. This relative transformation matrix can be used to determine the pose changes of objects in each frame of target point cloud data. Simultaneously, multiple target feature points corresponding to the target point cloud data can be converted into multiple first feature points in the same coordinate system. Based on these first feature points, a local map corresponding to that frame of target point cloud data can be constructed. Following the relative transformation matrix, these first feature points corresponding to multiple frames of point cloud data can be stitched together to obtain a complete high-precision map.

[0082] As an optional implementation, multiple target feature points corresponding to the target point cloud data are transformed to obtain a relative change matrix corresponding to the target point cloud data, including: determining second target point cloud data adjacent to the first target point cloud data, wherein the multi-frame target point cloud data includes the first target point cloud data and the second target point cloud data; in response to the average gradient corresponding to the first target feature point in the first target point cloud data being greater than the intensity gradient threshold, or the first target feature point being an edge feature point, determining a third target feature point adjacent to the first target feature point and a fourth target feature point next to the first target feature point among the multiple second target feature points corresponding to the second target point cloud data; constructing a line feature based on the third target feature point and the fourth target feature point; obtaining a first distance function between the line feature and the first target feature point; and constructing a relative change matrix based on the first distance function.

[0083] In this embodiment, multiple frames of target point cloud data can be processed sequentially. First, a second target point cloud data adjacent to the first target point cloud data can be determined, i.e., a reference frame adjacent to the current frame can be determined. If the average gradient corresponding to the first target feature point in the first target point cloud data is true, the first target feature point can be determined as an intensity feature point. If the first target feature point is an intensity feature point or an edge feature point, a third target feature point corresponding to the vector of the first target feature point and a fourth target feature point adjacent to the first target feature point can be determined from the feature points corresponding to the second target point cloud data. Based on the third and fourth target feature points, a line feature can be constructed. A first distance function between the line feature and the first target feature point is obtained, and based on the first distance function, a relative change matrix can be constructed.

[0084] Optionally, different processing methods can be used for different types of target feature points. For intensity feature points and edge feature points, a unified feature point matching framework can be used. Figure 2 This is a schematic diagram of edge feature point or intensity feature point matching according to an embodiment of this application, such as... Figure 2 As shown, for edge feature point or intensity feature point matching, we can first determine the edge point a (i.e., the first target feature point) of the current frame and the nearest neighbor point b (i.e., the third target feature point) of the reference frame through KD tree nearest neighbor search. Then, we obtain the second nearest neighbor point c (i.e., the fourth target feature point) on the adjacent scan line of point b, forming a line feature. By optimizing and minimizing the distance from point a to the line feature, we obtain the first distance function (de). Based on the first distance function, we can solve for the optimal pose transformation.

[0085] As an optional implementation, a relative change matrix is ​​constructed based on a first distance function, including: in response to the first target feature point being a planar feature point, determining, among multiple second target feature points corresponding to the second target point cloud data, a fifth target feature point adjacent to the first target feature point, a sixth target feature point adjacent to the first target feature point at the same scan line number, and a seventh target feature point adjacent to the scan line adjacent to the fifth target feature point; constructing a surface feature based on the fifth target feature point, the sixth target feature point, and the seventh target feature point; obtaining a second distance function between the surface feature and the second target feature point; and constructing a relative change matrix based on the first distance function and the second distance function.

[0086] In this embodiment, if the type of the first target feature point is a planar feature point, then among the multiple second target feature points corresponding to the second target point cloud data, a fifth target feature point adjacent to the first target feature point, a sixth target feature point adjacent to the first target feature point at the same scan line number, and a seventh target feature point adjacent to the scan line adjacent to the fifth target feature point can be determined; based on the fifth target feature point, the sixth target feature point, and the seventh target feature point, a surface feature can be constructed; a second distance function between the surface feature and the second target feature point is obtained; based on the first distance function and the second distance function, a relative change matrix can be constructed.

[0087] Figure 3 This is a schematic diagram of planar feature point matching according to an embodiment of this application, such as... Figure 3 As shown, for planar feature points, a triple nearest neighbor search strategy can be adopted. In the reference frame (i.e., the second target point cloud data), the nearest neighbor e (i.e., the fifth target feature point), the second nearest neighbor f (i.e., the sixth target feature point), and the nearest neighbor g (i.e., the seventh target feature point) of point d (i.e., the first target feature point) along the same scan line are selected to form the planar feature. By minimizing the distance from point d to the planar feature, the second distance function (df) is obtained. Based on the first and second distance functions, the relative transformation matrix can be constructed.

[0088] As an optional implementation, a relative change matrix is ​​constructed based on a first distance function and a second distance function, including: weighting and fusing the first distance function and the second distance function to obtain a target distance function; and performing nonlinear processing on the target distance function to obtain a relative change matrix.

[0089] In this embodiment, after obtaining the first distance function and the second distance function corresponding to multiple first target feature points using the above method, the first distance function and the second distance function can be weighted and fused to obtain the target distance function. The target distance function can be nonlinearly processed to obtain a relative change matrix, which can be used to characterize the pose change in each frame of the keyframe. Based on the pose change, multiple first feature points can be stitched together to obtain map data.

[0090] Optionally, weighted fusion of multiple distance functions can be performed using the Levenberg-Marquardt Algorithm, which nonlinearly minimizes multiple distance functions (i.e., multiple first distance functions and multiple second distance functions) to obtain a relative transformation matrix. Based on the relative transformation matrix, the pose change is determined, and based on the pose change, multiple first feature points are stitched together to obtain map data.

[0091] As an optional implementation, the method may further include: in response to the vehicle running on a target running path, acquiring the vehicle's target positioning information and the vehicle's running data, wherein the running data is used to characterize the vehicle's running state; constructing a graph model using multi-frame point cloud data, target positioning information, and running data, wherein the graph model includes multiple state nodes, the state nodes are used to characterize the vehicle's pose state at different times, and the connecting edges between state nodes are used to characterize the constraint relationships between state nodes; and using the graph model to determine the vehicle's position information in map data.

[0092] In this embodiment, the target positioning information can be GPS information. The operational data can be motion data collected by an Inertial Measurement Unit (IMU), which can be used to characterize the vehicle's operational state. The graphical model can include multiple state nodes, which can be used to characterize the vehicle's pose state at different times, such as the vehicle's position and orientation. The connecting edges between state nodes can be used to characterize the constraint relationships between state nodes, and can be factors, including but not limited to: IMU pre-integration factors, GNSS factors, laser odometry factors, and loop closure detection factors.

[0093] Optionally, a graph model can be constructed using multi-frame point cloud data, target positioning information, and operational data. Using this graph model, the vehicle's location information within the map data can be determined. This location information can then be matched with a high-precision map to determine whether the vehicle is traveling on the target path.

[0094] In this embodiment, a tightly coupled laser SLAM framework based on factor graph optimization is employed. By constructing a graph model containing state nodes and factors from multiple sensor sources, deep fusion of LiDAR point clouds, IMU motion data, and GPS positioning information is achieved. This framework transforms the maximum a posteriori probability estimation problem into a nonlinear least squares optimization problem. State nodes represent the temporal state of the vehicle's pose, and the factor set can include IMU pre-integration factors, laser odometry factors, GNSS factors, and loop closure detection factors. An incremental optimization strategy can be employed. When the vehicle pose change exceeds a dynamic threshold, a Bayesian tree incremental smoothing algorithm is triggered to perform an optimization calculation on the entire factor graph to update the vehicle's pose estimate at the current moment.

[0095] Optionally, for the IMU pre-integration factor, the high-frequency angular velocity and acceleration measurements can be converted into pose changes between adjacent state points to construct the IMU pre-integration factor to characterize continuous motion constraints. For the GNSS factor, an incremental correction mechanism with 5-meter intervals can be set to effectively suppress long-term drift while ensuring computational efficiency.

[0096] In this embodiment, a factor graph loop closure detection method based on Euclidean distance and the Iterative Closest Point algorithm (ICP) registration can also be used. This method constructs a spatial-temporal dual-threshold detection mechanism (a spatial distance range of 15 meters + a time interval window of 12 frames) to dynamically search for candidate loop closure points in the factor graph. The ICP algorithm is used to perform pose registration of the current frame with the sub-keyframes (a total of 25 frames) in the world coordinate system W, and the matching results are injected into the optimization process as loop closure factors.

[0097] Optionally, a high-precision map of the driving route is generated. For the constructed high-precision map, when the vehicle enters the route, the LiDAR acquires real-time point cloud data and performs ICP registration with the high-precision map. The registration accuracy is the confidence level. When the confidence level is greater than 95%, the route can be trusted as a frequently used driving route. The vehicle's pose estimation in the map can be obtained by transforming the point cloud data with the vehicle's pose.

[0098] The present application will be further explained and illustrated below with reference to specific embodiments.

[0099] Figure 4 This is a flowchart of a common driving route identification and optimization strategy according to an embodiment of this application, such as... Figure 4 As shown, the process may include the following steps.

[0100] Step S402: Extract features from the target point cloud data to obtain target feature points.

[0101] In this embodiment, in order to preliminarily determine whether a certain route is a frequently used driving route of a vehicle, GPS positioning information of the vehicle before driving and after parking within the past month is collected to obtain high-frequency positioning points, which are used as the starting point and ending point of the frequently used driving route of the vehicle.

[0102] Optionally, when the user travels the same route again, the SLAM algorithm is used to map the route scene and obtain a high-precision map of the scene. Intensity feature points are introduced to improve point cloud registration accuracy, building upon traditional edge and planar feature point extraction.

[0103] Optionally, the concept of intensity gradient is introduced for the extraction of intensity feature points. For each point in the point cloud to be extracted, a KD-tree nearest neighbor search is used with a search radius of 10 nearest neighbor points (excluding the current point itself). The searched neighbor indices are iterated, and the absolute difference between the intensity of the current point and the intensity of its neighbors is calculated. These differences are accumulated in the gradient variable, and the average gradient is the ratio of the accumulated gradient to the number of neighbor points. If the average gradient is greater than the intensity gradient threshold, the point is determined to be an intensity feature point. The calculation formula can be expressed as follows:

[0104]

[0105] Optionally, in large-scale environments, the extracted intensity feature points are mainly located at the edges where different materials intersect, and at points with complex surrounding environments. These points are usually located where the surface properties of objects change, such as the boundary between two materials with different reflectivities, or in areas where multiple structures intertwine in the scene. The curvature of the intensity feature points extracted from these areas may also meet the criteria for edge feature points, but replacing edge feature points with intensity feature points is unreasonable and meaningless. Therefore, intensity feature points also need to satisfy the following equation:

[0106]

[0107] Step S404: Match keyframes with local maps to output a high-precision map.

[0108] In this embodiment, the keyframe pose estimation method for lidar odometer based on a sliding window is characterized by: using a fixed-length sliding window to store lidar frames; determining the current frame as a keyframe when the vehicle pose change exceeds a translation threshold of 1.5m or a rotation threshold of 15°; performing pose estimation using a set of 25 consecutive keyframes on the time axis; establishing a set of transformation relationships between keyframes and the world coordinate system W; transforming each keyframe to the W coordinate system and stitching them together to generate a local map containing edge features, planar features, and intensity features; and performing downsampling processing on the stitched local map, where edge features and intensity features use a resolution of 0.2m, and planar features use a resolution of 0.4m, effectively eliminating redundant feature points and achieving lightweight real-time positioning and mapping.

[0109] Optionally, a unified feature point matching framework is used for both intensity and edge feature points. For matching edge and intensity feature points, firstly, a KD-tree nearest neighbor search is used to determine the nearest neighbor point b of the current frame's edge point a and the reference frame's edge point a. Then, the second nearest neighbor point c on the adjacent scan line of point b is obtained, forming a line feature. The distance function is obtained by optimizing and minimizing the distance from point a to the line feature, and the optimal pose transformation is solved.

[0110] Optionally, for planar feature points, a triple nearest neighbor search strategy is adopted: in the reference frame, the nearest neighbor e of point d, the second nearest neighbor f on the same scan line, and the nearest neighbor g on the adjacent scan line of point e are selected to form the planar feature. By minimizing the distance from point d to the planar feature, a distance function is obtained to achieve accurate matching of the planar feature.

[0111] Optionally, multiple distance functions are weighted and fused, and the Levenberg-Marquard algorithm is used to nonlinearly minimize the distance functions to obtain the relative transformation matrix. The pose change is determined based on the relative transformation matrix, and feature points are stitched together based on the pose change to obtain map data. This map data can be a high-precision map.

[0112] Step S406: Optimize the backend of the factor graph.

[0113] In this embodiment, a tightly coupled laser SLAM framework based on factor graph optimization is employed. By constructing a graph model containing state nodes and multi-source sensor factors, deep fusion of LiDAR point cloud, IMU motion data, and GPS positioning information is achieved. This framework transforms the maximum a posteriori probability estimation problem into a nonlinear least squares optimization problem. State nodes represent the temporal state of the vehicle's pose, and the factor set includes IMU pre-integration factors, laser odometry factors, GNSS factors, and loop closure detection factors. The system employs an incremental optimization strategy. When the vehicle pose change exceeds a dynamic threshold, a Bayesian tree incremental smoothing algorithm is triggered to perform an optimization calculation on the entire factor graph to update the vehicle's pose estimate at the current moment.

[0114] Optionally, for the IMU pre-integration factor, the high-frequency angular velocity and acceleration measurements are converted into pose changes between adjacent state points to construct the IMU pre-integration factor to characterize continuous motion constraints. For the GNSS factor, an incremental correction mechanism with 5-meter intervals is set to effectively suppress long-term drift while ensuring computational efficiency.

[0115] In this embodiment, a factor graph loop closure detection method based on Euclidean distance and ICP registration can also be used. This method constructs a spatial-temporal dual-threshold detection mechanism (a spatial distance range of 15 meters + a time interval window of 12 frames) to dynamically search for candidate loop closure points in the factor graph. The ICP algorithm is used to perform pose registration between the current frame and sub-keyframes (a total of 25 frames) in the world coordinate system W, and the matching results are injected into the optimization process as loop closure factors.

[0116] Optionally, a high-precision map of the driving route is generated. For the constructed high-precision map, when the vehicle enters the route, the LiDAR acquires real-time point cloud data and performs ICP registration with the high-precision map. The registration accuracy is the confidence level. When the confidence level is greater than 95%, the route can be trusted as a frequently used driving route. The vehicle's pose estimation in the map can be obtained by transforming the point cloud data with the vehicle's pose.

[0117] In this embodiment of the application, multi-frame point cloud data of a vehicle is acquired, and multi-frame target point cloud data in the multi-frame point cloud data is determined. The target point cloud data can be key frames in the multi-frame point cloud data. Based on the target feature points in the multi-frame key frames, map data can be constructed. The map data can be a high-precision map containing the vehicle's operating area, thereby achieving the technical effect of improving the accuracy of map construction and solving the technical problem of low accuracy in map construction.

[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0119] According to an embodiment of this application, a map construction apparatus embodiment is also provided. It should be noted that the apparatus can be used to run the map construction method described above.

[0120] Figure 5 This is a schematic diagram of a map construction apparatus according to an embodiment of this application. Figure 5 As shown, the map construction device may include: an acquisition unit 502, a filtering unit 504, a determination unit 506, and a construction unit 508.

[0121] The acquisition unit 502 is used to acquire multi-frame point cloud data of the vehicle in response to the vehicle running on the target running path. The point cloud data is used to represent the vehicle's pose state and the structural information of the map where the target running path is located.

[0122] The filtering unit 504 is used to filter out multiple frames of target point cloud data from multiple frames of point cloud data, wherein the pose state corresponding to the target point cloud data satisfies the target pose state.

[0123] The determining unit 506 is used to determine multiple target feature points from multiple feature points in the target point cloud data.

[0124] The construction unit 508 is used to construct map data for the obtained map based on multiple target feature points corresponding to multiple frames of target point cloud data.

[0125] In the map construction apparatus of this embodiment, an acquisition unit acquires multiple frames of point cloud data of the vehicle in response to the vehicle running on a target running path. The point cloud data is used to characterize the vehicle's pose state and the structural information of the map where the target running path is located. A filtering unit filters multiple frames of target point cloud data from the multiple frames of point cloud data, wherein the pose state corresponding to the target point cloud data satisfies the target pose state. A determination unit determines multiple target feature points from multiple feature points of the target point cloud data. A construction unit constructs map data based on the multiple target feature points corresponding to the multiple frames of target point cloud data, thereby achieving the technical effect of improving the performance detection efficiency of the operating system and solving the technical problem of low accuracy in map construction.

[0126] Embodiments of this application also provide a vehicle, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods described in various embodiments of this application.

[0127] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0128] Embodiments of this application also provide a computer program product, including a computer program that implements the methods of various embodiments of this application when run by a processor.

[0129] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program, which is executed by a processor to implement the methods in various embodiments of this application.

[0130] Embodiments of this application also provide a computer program that, when run by a processor, implements the methods described in the various embodiments of this application.

[0131] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0132] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not operational. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0133] The units described as separate components may or may not be physically separate. Similarly, the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0134] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0135] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to run all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0136] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a map, characterized in that, include: In response to the vehicle running on the target running path, multi-frame point cloud data of the vehicle is acquired, wherein the point cloud data is used to characterize the pose state of the vehicle and the structural information of the map where the target running path is located. From the multiple frames of point cloud data, multiple frames of target point cloud data are selected, wherein the pose state corresponding to the target point cloud data satisfies the target pose state. Multiple target feature points are determined from multiple feature points in the target point cloud data; Map data for the map is constructed based on multiple target feature points corresponding to multiple frames of target point cloud data.

2. The method according to claim 1, characterized in that, The step of filtering out multiple frames of target point cloud data from multiple frames of point cloud data includes: Based on the point cloud data, the pose state of the vehicle is determined; In response to the pose state satisfying the target pose state, the point cloud data is determined as the target point cloud data, wherein the target pose state is used to characterize the state in which the pose change of the vehicle exceeds an adjustment threshold.

3. The method according to claim 1, characterized in that, The step of determining multiple target feature points from multiple feature points in the target point cloud data includes: The average gradients corresponding to the plurality of feature points are determined respectively, and at least one edge feature point and at least one planar feature point are determined among the plurality of feature points, wherein the average gradients are used to characterize the intensity change characteristics of the feature points; In response to the average gradient being greater than the intensity gradient threshold, the feature point is determined as the target feature point, and the edge feature point or the planar feature point is also determined as the target feature point.

4. The method according to claim 3, characterized in that, The map data, constructed based on multiple target feature points corresponding to multiple frames of target point cloud data, includes: The target point cloud data is transformed to obtain a relative transformation matrix corresponding to the target point cloud data, wherein the relative transformation matrix is ​​used to characterize the pose change state of the target feature points; Multiple target feature points corresponding to multiple frames of target point cloud data are converted into multiple first feature points in the same coordinate system, wherein the first feature points correspond to the target feature points; The map data is obtained by stitching together multiple first feature points using the relative change matrix.

5. The method according to claim 4, characterized in that, The step of transforming multiple target feature points corresponding to the target point cloud data to obtain the relative change matrix corresponding to the target point cloud data includes: A second target point cloud data adjacent to the first target point cloud data is determined, wherein the target point cloud data in multiple frames includes the first target point cloud data and the second target point cloud data; In response to the average gradient corresponding to the first target feature point in the first target point cloud data being greater than the intensity gradient threshold, or the first target feature point being the edge feature point, a third target feature point adjacent to the first target feature point and a fourth target feature point adjacent to the first target feature point are determined among a plurality of second target feature points corresponding to the second target point cloud data. Based on the third target feature point and the fourth target feature point, a line feature is constructed; Obtain the first distance function between the line feature and the first target feature point; The relative change matrix is ​​constructed based on the first distance function.

6. The method according to claim 5, characterized in that, The process of constructing the relative change matrix based on the first distance function includes: In response to the first target feature point being the planar feature point, among the multiple second target feature points corresponding to the second target point cloud data, a fifth target feature point adjacent to the first target feature point, a sixth target feature point adjacent to the first target feature point at the same scan line number, and a seventh target feature point adjacent to the scan line adjacent to the fifth target feature point are determined. Based on the fifth target feature point, the sixth target feature point, and the seventh target feature point, a surface feature is constructed; Obtain the second distance function between the surface feature and the second target feature point; The relative change matrix is ​​constructed based on the first distance function and the second distance function.

7. The method according to claim 6, characterized in that, The process of constructing the relative change matrix based on the first distance function and the second distance function includes: The first distance function and the second distance function are weighted and fused to obtain the target distance function; The relative change matrix is ​​obtained by performing nonlinear processing on the target distance function.

8. The method according to any one of claims 1 to 7, characterized in that, The method further includes: In response to the vehicle operating on the target operating path, the target positioning information of the vehicle and the operating data of the vehicle are obtained, wherein the operating data is used to characterize the operating status of the vehicle; Using the point cloud data from multiple frames, the target positioning information, and the running data, a graph model is constructed. The graph model includes multiple state nodes, which represent the pose state of the vehicle at different times. The connecting edges between the state nodes represent the constraint relationships between the state nodes. Using the graph model, the location information of the vehicle in the map data is determined.

9. A vehicle, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method of any one of claims 1 to 8.

Images