A method and system for generating two-dimensional raster maps

Abstract

This application provides a method and system for generating two-dimensional grid maps. The generation method includes: acquiring a set of keyframe poses and a set of keyframe point clouds output by a SLAM system; traversing the set of keyframe point clouds and performing the following operations on each keyframe point cloud: transforming the point cloud to the world coordinate system using the corresponding keyframe poses in the set of keyframe poses; performing ray penetration statistics: acquiring discrete points; mapping the discrete points to corresponding grids on the initial two-dimensional grid map; updating the statistical values ​​of the grids traversed between the ray start point and the end point to characterize drivability evidence; updating the statistical values ​​of the grid where the end point is located to characterize obstacle evidence; classifying each grid according to the final statistical value of each grid in the initial two-dimensional grid map; and outputting a target two-dimensional grid map to display the classification results. This application achieves accurate classification of grid maps through a ray penetration statistics mechanism.

Description

Technical Field

[0001] This invention relates to the technical field of environmental perception for autonomous driving, and specifically to a method and system for generating two-dimensional grid maps. Background Technology

[0002] Simultaneous localization and mapping (SLAM) based on 3D LiDAR is widely used in environmental modeling for various intelligent autonomous vehicles due to its high-precision environmental geometry detection capabilities. However, due to interference from dynamic objects in the environment, the 3D point cloud map it constructs cannot directly serve the environmental perception and motion planning of autonomous vehicles.

[0003] The current mainstream approach is to convert 3D point cloud maps into 2D raster maps to adapt to the operational characteristics of autonomous vehicles in ground environments. Existing raster map conversion schemes often incorporate point cloud map boundary features to create the raster map. However, in real-world operating environments, the resulting raster maps cannot accurately distinguish between drivable, non-drivable, and unknown areas. For example, related technical solutions typically use point cloud map boundary features to create raster maps, which can only identify obstacle boundaries and cannot effectively distinguish between undetected and drivable areas. This results in incomplete environmental perception information, which in turn affects the autonomous vehicle's subsequent motion planning.

[0004] Therefore, there is an urgent need for a method and system for generating two-dimensional raster maps to meet the needs of practical applications. Summary of the Invention

[0005] The main objective of this invention is to provide a method and system for generating two-dimensional raster maps, in order to overcome the shortcomings of related technologies.

[0006] The objective of this application is achieved through the following technical solution: Firstly, this application provides a method for generating a two-dimensional raster map, including: Obtain the keyframe pose set and keyframe point cloud set output by the SLAM system; iterate through the keyframe point cloud set and perform the following operations on each keyframe point cloud: Using the keyframe poses corresponding to the keyframe pose set, the point cloud is transformed into the world coordinate system; For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids traversed between the ray's starting point and ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. After traversing all keyframe point clouds, the grids in the initial two-dimensional grid map are classified according to their final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas; a target two-dimensional grid map is output to display the classification results.

[0007] In some optional embodiments, the initial two-dimensional raster map generation step includes: Based on the boundaries of the prior point cloud map and the preset raster map resolution, the size and origin of the initial two-dimensional raster map are determined, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

[0008] In some optional embodiments, updating the statistics to characterize the evidence of accessibility includes: decrementing the statistics of the corresponding grid by 1; updating the statistics to characterize the evidence of obstacles includes: incrementing the statistics of the corresponding grid by 1.

[0009] In some optional embodiments, classifying areas based on final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas includes: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

[0010] In some alternative embodiments, before transforming the point cloud to the world coordinate system, the method further includes: The keyframe point cloud is height filtered to remove points whose height is below a preset ground threshold or above a preset vehicle top threshold, and retains point clouds whose height is within the driving height range of the unmanned vehicle.

[0011] In some optional embodiments, the projection of the ray onto a two-dimensional plane and the discretization sampling specifically involves: The ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

[0012] Secondly, this application provides a system for generating two-dimensional raster maps, comprising: The collection acquisition and processing module is used to acquire the keyframe pose set and keyframe point cloud set output by the SLAM system; it iterates through the keyframe point cloud set and performs the following operations on each keyframe point cloud: The point cloud is transformed into the world coordinate system using the corresponding keyframe poses in the keyframe pose set. For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids that the ray passes through from the starting point to the ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. The classification and display module is used to classify each grid in the initial two-dimensional grid map according to the final statistical value of each grid after traversing all keyframe point clouds, so as to distinguish between drivable areas, non-drivable areas and unknown areas; and outputs a target two-dimensional grid map to display the classification results.

[0013] In some optional embodiments, the initial two-dimensional raster map generation step includes: Based on the boundaries of the prior point cloud map and the preset raster map resolution, the size and origin of the initial two-dimensional raster map are determined, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

[0014] In some optional embodiments, updating the statistics to characterize the evidence of drivability includes: decrementing the statistics of the corresponding grid by 1; updating the statistics to characterize the evidence of obstacles includes: incrementing the statistics of the corresponding grid by 1. The classification based on the final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas includes: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

[0015] In some optional embodiments, the projection of the ray onto a two-dimensional plane and the discretization sampling specifically involves: The ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

[0016] Compared with related technologies, the advantages of this invention are as follows: By utilizing the ray penetration statistical mechanism, it fully leverages the keyframe information of the 3D point cloud map constructed during SLAM, converting the ray information in the keyframes into statistical values ​​of the grid map. Through the three states of positive, negative, and zero statistical values, it accurately distinguishes drivable, non-drivable, and unknown areas. Compared to related technologies that only use point cloud boundary features, this method more accurately reflects the true information of the drivable area of ​​the autonomous vehicle. Through height filtering, points with heights below a preset ground threshold or above a preset vehicle top threshold are removed, effectively eliminating interference from ground points and dynamic objects above the vehicle, avoiding the problem of misclassifying drivable areas as obstacle areas, and improving the accuracy of the grid map. Employing a ray hit and penetration statistical method, for grids traversed between the ray's origin and destination, the statistical value is decremented by 1 to represent drivability evidence; for the grid where the ray ends, the statistical value is incremented by 1 to represent obstacle evidence. This statistical method is logically clear, computationally simple, easy to implement, and highly accurate. The generated two-dimensional grid map can accurately distinguish between drivable areas, non-drivable areas, and unknown areas, providing reliable basic data for the environmental perception and motion planning of autonomous vehicles, and effectively improving the autonomous navigation capabilities of autonomous vehicles in complex environments. Attached Figure Description

[0017] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0018] Figure 1 This is a flowchart illustrating a method for generating a two-dimensional raster map according to an embodiment of this application; Figure 2 This is a schematic diagram of a target two-dimensional grid map provided in an embodiment of this application; Figure 3 This is a top view of a three-dimensional point cloud provided in an embodiment of this application; Figure 4 This is a side view of a three-dimensional point cloud provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a two-dimensional raster map generation system provided in an embodiment of this application. Detailed Implementation

[0019] The present application will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0020] Taking, for example, the method, system, terminal, and medium for generating a grid map of a mining area for unmanned mining trucks provided in application number CN202511713004.9, the following approach: First, static environmental data of the loading area is collected by multi-source sensors deployed on the mining truck. After denoising, time synchronization, and graph-optimized SLAM fusion processing, an initial point cloud map is generated. Then, through automated point cloud rasterization classification and boundary extraction, the initial grid map is exported. Based on this, a real-time update mechanism is established: by continuously collecting dynamic operation data, the current loading boundary features are extracted and compared geometrically with the initial boundary to determine area changes. If the change exceeds a threshold, an incremental update process is triggered, performing local recalculation and map overlay only on the changed area, ultimately outputting the updated grid map. This approach can achieve high-precision automatic construction and efficient real-time maintenance of mining area grid maps, improving the safety and efficiency of unmanned mining truck operations. However, the generated grid map cannot reflect the true information of the drivable area of ​​the unmanned vehicle. Furthermore, it does not consider the height limitations of the unmanned vehicle itself and interference from dynamic objects.

[0021] In other words, existing methods for converting 3D point clouds into 2D grid maps cannot reflect the true information of the drivable area of ​​autonomous vehicles and affect their subsequent motion planning. This is because related technical solutions typically introduce point cloud map boundary features to draw grid maps, failing to consider the influence of dynamic objects during the 3D point cloud map construction process. Furthermore, they neglect the autonomous vehicle's own height information, causing the generated grid map to classify some drivable areas as obstacle areas, resulting in distorted environmental perception.

[0022] Based on this, this application provides a method and system for generating a two-dimensional grid map. It fully utilizes keyframe information from the 3D point cloud ray penetration statistics to construct the 3D point cloud map during the SLAM process. After determining the map boundary information, the grid map is initialized. Then, the ray information of each point in each keyframe is calculated, and the number of times each grid cell is hit and penetrated is counted. Finally, based on the grid penetration statistics, the classification of the grid cells (drivable, non-drivable, unknown) is determined, generating an accurate and reliable two-dimensional grid map. During grid map initialization, point cloud information within the robot's height range of the keyframes can be statistically analyzed. The method will be described first, followed by the system.

[0023] Example 1 This application provides a method for generating a two-dimensional grid map. This method is based on a three-dimensional point cloud ray penetration statistical mechanism and can accurately distinguish between drivable areas, non-drivable areas, and unknown areas.

[0024] See Figure 1 The method for generating a two-dimensional raster map includes the following steps: Step S101: Obtain the keyframe pose set and keyframe point cloud set output by the SLAM system.

[0025] Specifically, during the operation of the SLAM system, a 3D point cloud map is constructed and keyframe information, including a set of keyframe poses and a set of keyframe point clouds, is output. In this embodiment, the coordinate systems of the robot body and the world are defined as B and W, respectively, and the coordinate system of the LiDAR coincides with the coordinate system of the robot body. The robot's pose state is described as follows during a LiDAR scan: ,in It is the rotation matrix of the robot's posture in the W series. It is the corresponding translation vector.

[0026] Step S102: Traverse the keyframe point cloud set and perform coordinate transformation and ray penetration statistics operations on each keyframe point cloud.

[0027] Specifically, the following operations are performed on each keyframe point cloud: Using the keyframe poses corresponding to the keyframe pose set, the point cloud is transformed into the world coordinate system; For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids traversed between the ray's starting point and ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. Step S103: After traversing all keyframe point clouds, classify each grid in the initial two-dimensional grid map according to the final statistical value of each grid to distinguish between drivable areas, non-drivable areas and unknown areas; output the target two-dimensional grid map to display the classification results.

[0028] For the grid cells traversed between the ray's origin and destination (i.e., the grid cells the ray penetrates), their statistical values ​​are updated to characterize the evidence of traversability (in practice, the statistical value is usually decremented by 1). The technical logic is that if the ray can reach subsequent points, it proves that there are no obstacles hindering the propagation of the laser within the grid space, making it a traversable space.

[0029] For each grid cell where the ray terminates (i.e., the cell hit by the ray), its statistical value is updated to characterize the obstacle evidence (in practice, the statistical value is usually incremented by 1). The technical logic is that the ray being blocked at this point indicates the presence of a physical obstacle within that grid cell.

[0030] This embodiment utilizes the keyframe information for constructing a 3D point cloud map during the SLAM process through a ray penetration statistics mechanism, achieving accurate classification of the raster map and effectively distinguishing between drivable areas, non-drivable areas, and unknown areas.

[0031] Example 2 This embodiment, based on Embodiment 1, provides a detailed explanation of the specific implementation methods for generating the initial two-dimensional raster map, height filtering, and ray projection discretization.

[0032] As one implementation method, the initial two-dimensional raster map generation step includes: determining the size and origin of the initial two-dimensional raster map based on the boundary of the prior point cloud map and the preset raster map resolution, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

[0033] Specifically, the prior point cloud map is constructed using the LIO-SAM method. This prior point cloud map is composed of stitched-together point clouds from several keyframes. There are a total of multiple keyframes, and the keyframe pose set and the B-frame point cloud set are respectively... and .

[0034] The LIO-SAM method is exemplified by "LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping," which was included in the 2020 IEEE / RSJ International Conference on Intelligent Robots and Systems (IROS). doi:10.1109 / IROS45743.2020.9341176.

[0035] As one implementation method, before converting the point cloud to the world coordinate system, the method further includes: performing height filtering on the keyframe point cloud, removing points whose height is lower than a preset ground threshold or higher than a preset vehicle top threshold, and retaining point clouds whose height is within the driving height range of the unmanned vehicle.

[0036] Specifically, point cloud filtering is performed based on the vertical height range of the autonomous vehicle relative to the installed LiDAR, eliminating invalid point clouds outside the vehicle's operating height range. This effectively removes interfering points on the ground and above the vehicle, preventing drivable areas from being misclassified as obstacle areas.

[0037] As one implementation method, the ray is projected onto a two-dimensional plane and discretized. Specifically, the ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

[0038] Specifically, the ray r is projected onto the xy plane to determine the starting and ending grid cells of the ray. The distance parameter in the ray formula is adjusted using a fixed step size (e.g., 0.01). The number of grid cells is gradually increased to count the other cells traversed by the start and end points. It should be understood that the step size can be adjusted according to actual needs; a smaller step size results in more accurate statistics, but also increases the computational load. The ray formula is: start point + direction vector × distance parameter, where the distance parameter ranges from (0, ∞).

[0039] Example 3 This embodiment, based on Embodiments 1 and 2, provides a detailed explanation of the statistical value update method and raster classification rules.

[0040] As one implementation, updating the statistical value to characterize accessibility evidence includes: decrementing the statistical value of the corresponding grid by 1; updating the statistical value to characterize obstacle evidence includes: incrementing the statistical value of the corresponding grid by 1.

[0041] Specifically, for each grid cell traversed between the ray's starting and ending points, a penetration is identified, and its statistical value is decremented by 1; for each grid cell at the endpoint, a hit is identified, and its statistical value is incremented by 1. This statistical method is logically clear, computationally simple, easy to implement, and highly accurate.

[0042] As one implementation method, areas are classified based on final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas, including: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

[0043] The classification rule is based on the following principle: if a grid cell is penetrated multiple times but rarely hit, it indicates that there are no obstacles in the area, the statistical value tends to be negative, and it should be marked as a drivable area; if a grid cell is hit multiple times, it indicates that there are obstacles in the area, the statistical value tends to be positive, and it should be marked as a non-drivable area; if a grid cell is neither penetrated nor hit, the statistical value remains at 0, it indicates that the area has not been detected and should be marked as an unknown area.

[0044] To facilitate understanding of Embodiments 1 to 3 provided in this application, a specific application embodiment of the technical solution of this application is provided below to illustrate the application effect of the above method in a real-world scenario.

[0045] like Figure 2 , Figure 3 and Figure 4 As shown, Figure 2The diagram shows the final output of a two-dimensional grid map based on the method of the present invention. White areas represent drivable areas, gray areas represent unknown areas, and black areas represent non-drivable areas. Figure 3 and Figure 4 These are the top view and side view of the corresponding 3D point cloud.

[0046] In an unmanned inspection robot scenario in an industrial park, the method of this invention is used to generate a grid map of the park's roads. A LiDAR is mounted on top of the inspection robot. A height filtering range is set to [-0.5m, 2.0m] to eliminate interference from ground points and points above the vehicle. The height filtering range is set based on the vehicle's coordinate system, not the world coordinate system; therefore, this threshold is related to the installation height of the LiDAR within the vehicle.

[0047] The specific solution involves defining the robot's coordinate system and the world's coordinate system as B and W respectively, with the lidar coordinate system coinciding with the robot's coordinate system. The robot's pose state is described as follows during a LiDAR scan: ,in It is the rotation matrix of the robot's posture in the W series. This is the corresponding translation vector. The prior point cloud map is constructed using the LIO-SAM method. This prior point cloud map is composed of several keyframe point clouds stitched together. There are n (128) keyframes in total, and their keyframe pose sets and B-frame point cloud sets are respectively... and Based on the existing information, a two-dimensional raster map can be initialized. , The maxima and minima of the prior point cloud map in the xy plane are used to determine the size range of the raster map. s represents the raster with index (i,j) in the raster map, where i is the horizontal index and j is the vertical index; s is the set raster map resolution (the raster map resolution is set to 0.2m). It is an index The grid is set to a statistical value of 0, indicating it is unknown. Areas that have not been hit are marked as unknown rather than drivable because the map has a maximum boundary value, so there will inevitably be a large number of unhidden areas. For vehicle safety reasons, vehicles are only allowed to operate in drivable areas, and unknown areas are treated the same as non-drivable areas.

[0048] Traversal The point cloud information is used to determine the vertical height range of the autonomous vehicle relative to the installed LiDAR. Point cloud filtering is performed to remove invalid point clouds within the operating altitude range of non-autonomous vehicles; the filtered point cloud is then used. For example, transferring it to the W series yields... traversal Each point in The ray from the keyframe position to the valid point is obtained, which can be represented as: ,in Project the ray onto the xy plane to determine the ray. starting grid With the endpoint grid Make it with a step size of 0.01 The count of other grid cells passed through the start and end points is gradually increased. A grid cell that has been passed through is considered to have been breached, and its count is decreased by 1. The end grid cell is considered to have been hit, and its count is increased by 1.

[0049] Traversing a 2D raster map For each grid cell, observe its statistical value. If the statistical value is equal to 0, it is considered an unknown area; if the statistical value is less than 0, it is considered a drivable area; if the statistical value is greater than 0, it is considered a non-drivable area. Finally, the two-dimensional grid map is output.

[0050] In addition, in practical applications, the weight of the rays passing through the grid can be differentiated, instead of treating them all the same by adding or subtracting 1, and instead judging them based on the distance the ray travels through the grid.

[0051] For example, s is the set grid map resolution. The starting and ending points of a ray passing through the grid are calculated, and the distance d it travels within the grid is obtained. Its weight is min(1.0, d / s). When d is not less than s, the weight is 1, meaning the full weight of the original scheme is applied; when d is less than s, the weight is less than 1, and the ray's contribution is attenuated. The weights for hits and penetrations are calculated as follows, replacing the original increment / decrement operation. In this case, the grid no longer only records integer vote counts, but records weighted cumulative weight data. The longer the ray travels, the greater the weight (positive correlation). In this case, noise from short rays is naturally suppressed.

[0052] Compared with related technologies, the method of this invention can effectively avoid misjudging drivable areas as obstacle areas, improve the accuracy of environmental perception, and provide reliable basic data for autonomous navigation of unmanned vehicles.

[0053] Therefore, the technical solution provided in this application makes full use of the key frame information for building a three-dimensional point cloud map during the SLAM process. Under the premise of filtering the drivable area of ​​the unmanned vehicle, the occupancy of each grid is counted by ray hit and penetration, and finally a precise two-dimensional grid map is output.

[0054] Example 4 This embodiment provides a two-dimensional raster map generation system for implementing the methods described in embodiments 1-3 above. Its specific effects are the same as those in the method embodiments, and will not be repeated here.

[0055] like Figure 5 As shown, the system includes: The collection acquisition and processing module is used to acquire the keyframe pose set and keyframe point cloud set output by the SLAM system; it iterates through the keyframe point cloud set and performs the following operations on each keyframe point cloud: The point cloud is transformed into the world coordinate system using the corresponding keyframe poses in the keyframe pose set. For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids that the ray passes through from the starting point to the ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. The classification and display module is used to classify each grid in the initial two-dimensional grid map according to the final statistical value of each grid after traversing all keyframe point clouds, so as to distinguish between drivable areas, non-drivable areas and unknown areas; and outputs a target two-dimensional grid map to display the classification results.

[0056] As one implementation method, the initial two-dimensional raster map generation step includes: determining the size and origin of the initial two-dimensional raster map based on the boundary of the prior point cloud map and the preset raster map resolution, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

[0057] As one implementation, updating the statistical value to characterize the evidence of accessibility includes: decrementing the statistical value of the corresponding grid by 1; updating the statistical value to characterize the evidence of obstacles includes: incrementing the statistical value of the corresponding grid by 1. They are categorized based on the final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas, including: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

[0058] As one implementation method, the ray is projected onto a two-dimensional plane and discretized. Specifically, the ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

[0059] As one implementation, it also includes a point cloud processing module, used to perform height filtering on the keyframe point cloud, remove points whose height is lower than a preset ground threshold or higher than a preset vehicle top threshold, and retain point clouds whose height is within the driving height range of the unmanned vehicle.

[0060] The terms “first,” “second,” “third,” “fourth,” “fifth,” “sixth,” “seventh,” “eighth,” “ninth,” etc. (if present) 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, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover 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.

[0061] This application describes the invention from the perspectives of purpose, performance, progress, and novelty, and it meets the functional enhancement and use requirements emphasized by the Patent Law. The above description and drawings are merely preferred embodiments of this application and are not intended to limit this application. Therefore, all structures, devices, features, etc., that are similar to or identical to those of this application, i.e., all equivalent substitutions or modifications made in accordance with the scope of this patent application, shall fall within the scope of protection of this patent application.

Claims

1. A method for generating a two-dimensional raster map, characterized in that, include: Obtain the keyframe pose set and keyframe point cloud set output by the SLAM system; iterate through the keyframe point cloud set and perform the following operations on each keyframe point cloud: Using the keyframe poses corresponding to the keyframe pose set, the point cloud is transformed into the world coordinate system; For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids traversed between the ray's starting point and ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. After traversing all keyframe point clouds, the points are classified according to the final statistical values ​​of each grid in the initial two-dimensional grid map to distinguish between drivable areas, non-drivable areas, and unknown areas. Output a target two-dimensional raster map to display the classification results.

2. The generation method according to claim 1, characterized in that, The steps for generating the initial two-dimensional raster map include: Based on the boundaries of the prior point cloud map and the preset raster map resolution, the size and origin of the initial two-dimensional raster map are determined, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

3. The generation method according to claim 1, characterized in that, The step of updating the statistical value to characterize the evidence of accessibility includes: decrementing the statistical value of the corresponding grid by 1; the step of updating the statistical value to characterize the evidence of obstacles includes: incrementing the statistical value of the corresponding grid by 1.

4. The generation method according to claim 3, characterized in that, The classification based on the final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas includes: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

5. The generation method according to claim 1, characterized in that, Prior to transforming the point cloud to the world coordinate system, the method further includes: The keyframe point cloud is height filtered to remove points whose height is below a preset ground threshold or above a preset vehicle top threshold, and retains point clouds whose height is within the driving height range of the unmanned vehicle.

6. The generation method according to claim 1, characterized in that, The process of projecting the ray onto a two-dimensional plane and discretizing it specifically involves: The ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

7. A system for generating a two-dimensional raster map, characterized in that, include: The collection acquisition and processing module is used to acquire the keyframe pose set and keyframe point cloud set output by the SLAM system; Iterate through the keyframe point cloud set and perform the following operations on each keyframe point cloud: The point cloud is transformed into the world coordinate system using the corresponding keyframe poses in the keyframe pose set. For each point in the transformed point cloud, a ray penetration statistics operation is performed: a ray is constructed from the sensor position of the keyframe to the point, the ray is projected onto a two-dimensional plane and discretized to obtain discrete points; the discrete points are mapped to the corresponding grids on the initial two-dimensional grid map; for the grids that the ray passes through from the starting point to the ending point, their statistical values ​​are updated to characterize accessibility evidence; for the grid where the ending point is located, its statistical values ​​are updated to characterize obstacle evidence. The classification and display module is used to classify each grid in the initial two-dimensional grid map according to the final statistical value of each grid after traversing all keyframe point clouds, so as to distinguish between drivable areas, non-drivable areas and unknown areas. Output a target two-dimensional raster map to display the classification results.

8. The generation system according to claim 7, characterized in that, The steps for generating the initial two-dimensional raster map include: Based on the boundaries of the prior point cloud map and the preset raster map resolution, the size and origin of the initial two-dimensional raster map are determined, wherein the initial statistical value of each raster of the initial two-dimensional raster map is 0.

9. The generation system according to claim 7, characterized in that, The step of updating the statistical value to characterize the evidence of accessibility includes: decrementing the statistical value of the corresponding grid by 1; the step of updating the statistical value to characterize the evidence of obstacles includes: incrementing the statistical value of the corresponding grid by 1. The classification based on the final statistical values ​​to distinguish between drivable areas, non-drivable areas, and unknown areas includes: If the statistical value of a raster is equal to 0, it is marked as an unknown area; If the statistical value of a grid is less than 0, it is marked as a drivable area; If the grid's statistical value is greater than 0, it is marked as a non-driveable area.

10. The generation system according to claim 7, characterized in that, The process of projecting the ray onto a two-dimensional plane and discretizing it specifically involves: The ray is projected onto a horizontal plane to obtain a two-dimensional ray; the parameters of the two-dimensional ray are sampled with a fixed step size to obtain a series of two-dimensional discrete points, which are used to map onto a grid.

Images