Point cloud slice-based grid map generation method, device, equipment and medium

Abstract

The application provides a point cloud layer-based grid map generation method, device, equipment and medium, which comprises the following steps: obtaining three-dimensional point cloud data of a target scene, mapping the horizontal plane coordinates to a grid, taking the lowest height value of the point cloud in each grid as the local ground height; offsetting the first preset height value and the second preset height value corresponding to the grid above the local ground height of each grid to form a collision zone, and only retaining the point cloud with a height located in the collision zone as a candidate point; for the grid whose number of candidate points reaches a preset threshold, marking it as an obstacle grid, otherwise marking it as a passable grid or a no-data grid, and generating an occupancy grid map. The application realizes high-precision and high-reliability automatic generation from point cloud to grid map.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus, device and medium for generating raster maps based on point cloud slicing. Background Technology

[0002] 3D point cloud map processing is a key component of robot indoor navigation technology. Its core challenge lies in how to accurately extract passable areas from massive point cloud data and construct reliable grid maps.

[0003] Existing solutions acquire 3D point clouds and then construct 3D dynamic grids and occupancy maps. Accuracy is enhanced through multi-source sensor fusion, and safety redundancy is increased based on device voltage level differentiation and device health status to achieve collision detection and path planning. Another solution first generates a 3D occupancy grid and then converts it to a 2D occupancy grid. During the 3D-to-2D conversion, ground point clouds at a certain height need to be deleted, and multi-layer grid fusion is used to process point clouds at different heights.

[0004] However, the shortcomings of the existing solutions mentioned above are: Existing technologies mostly use fixed height thresholds or global plane fitting to detect the ground. Indoor ceilings are typically two to three and a half meters high, and the maximum Z value of almost all XY grids is pushed to over three meters by the ceiling, causing the traditional maximum Z value thresholding method to misclassify the ground as an obstacle. Global fitting methods also struggle to accurately segment complex terrains such as slopes. Furthermore, grid generation and path planning are disconnected; rasterization may use coarse sampling data from the LOD (Level of Detail) parent layer, resulting in ghost obstacles or missing real obstacles.

[0005] In conclusion, there is an urgent need for a high-precision and high-reliability raster map generation method to solve the above problems. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this application provides a method, apparatus, device, and medium for generating raster maps based on point cloud slicing, thereby improving the accuracy and reliability of raster map generation.

[0007] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides a method for generating a raster map based on point cloud slices, including: Acquire 3D point cloud data of the target scene, map it to a grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height; Offset the local ground height of each grid by a first preset height value and a second preset height value corresponding to the grid to form a collision zone, and retain only the point cloud with a height within the collision zone as candidate points; For grids where the number of candidate points reaches a preset threshold, they are marked as obstacle grids; otherwise, they are marked as passable grids or grids with no data, and an occupied grid map is generated.

[0008] The beneficial effects of this application are as follows: by independently calculating the local ground height in each grid and offsetting it to form a collision zone, the ceiling point cloud naturally falls above the collision zone and is excluded, while the ground noise point cloud falls below the collision zone and is filtered out, fundamentally eliminating the interference of indoor ceilings on obstacle detection; by determining the number of candidate points by a threshold, the real obstacle grid is distinguished from the passable grid, forming a reliable occupancy grid map, realizing high-precision and high-reliability automated generation from point cloud to grid map.

[0009] Optionally, after acquiring the 3D point cloud data of the target scene, the method further includes: The node tree of the point cloud tile index is recursively parsed, and the point cloud files corresponding to the leaf nodes without child nodes are extracted as the point cloud data to participate in the raster mapping.

[0010] Optionally, the first preset height value is a preset ground thickness value; the second preset height value is a preset ceiling height value; the local ground height offset from the first preset height value is used as the lower boundary of the collision zone; and the local ground height offset from the second preset height value is used as the upper boundary of the collision zone.

[0011] Optionally, the process of generating the preset threshold includes: A preset threshold AMOP is dynamically calculated based on the global average number of points per grid in the 3D point cloud data. AMOP = max(int(avg×a), b); In the formula, avg is the global average number of points per grid in the three-dimensional point cloud data, int is the floor function, max is the maximum value function, a is the preset scaling factor, and b is the preset lower limit protection value.

[0012] Optionally, after generating the occupied raster map, the method further includes: Morphological opening is performed on the occupied grid map. First, the erosion rule is executed, and then the dilation rule is executed. The erosion rule is that if the number of obstacle cells in the eight-neighborhood of an obstacle cell is less than a preset degradation threshold, then the obstacle cell is downgraded to a passable cell. The dilation rule is that if the number of obstacle cells in the eight-neighborhood of a passable cell reaches a preset recovery threshold, then the passable cell is restored to an obstacle cell. The calculated occupied grid map is filtered by connected component area. For obstacle connected components with physical areas smaller than a preset area threshold, they are filtered and removed based on physical area, point cloud density, height, and neighborhood continuity. The remaining obstacle connected components are retained in the occupied grid map as real obstacle areas.

[0013] Optionally, after generating the occupied raster map, the method further includes: Obtain manually annotated semantic polygons; For semantic polygons of type wall, set the grids within the coverage area of ​​the semantic polygon to be impassable; For semantic polygons of type risk region, the passage cost of the raster within the coverage area of ​​the semantic polygon is superimposed with semantic risk weight; For impassable graticles, perform distance cost inflation, and apply a decay cost to surrounding graticles based on the Chebyshev distance: cost = c × (1 - (cheb-1) / R); R = ceil(d / r); In the formula, c is the preset attenuation coefficient, cheb is the Chebyshev distance from the current grid to the nearest impassable grid, R is the preset expansion radius, ceil is the round-up function, d is the preset safety distance constant, and r is the grid resolution.

[0014] Optionally, after generating the occupied raster map, the method further includes: A heuristic search algorithm is used to search for paths on an occupancy raster map with semantic risk weights superimposed. The heuristic function is the Octile distance h. h=max(|dx|,|dy|)+(√2-1)×min(|dx|,|dy|); In the formula, dx and dy are the grid distances from the current node to the target node in two orthogonal directions on the horizontal plane, respectively; The searched path is streamlined by using the Bresenham line detection algorithm to check the straightness between path nodes. If there is no grid with a cost greater than 0 between two nodes, the intermediate redundant node is deleted. The path after the line-of-sight shortcut is simplified using Douglas-Peucker simplification, outputting a simplified navigation waypoint sequence and calculating the heading angle of each waypoint. The heading angle of the intermediate waypoint is taken from the tangent direction of the line connecting the preceding and following waypoints, while the heading angles of the first and last waypoints are specified by the user or taken from the direction of the adjacent waypoint.

[0015] Secondly, this application provides a raster map generation device based on point cloud slicing layers, comprising: The point cloud acquisition module is used to acquire 3D point cloud data of the target scene, map it to a grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height. The collision zone construction module is used to offset the local ground height of each grid by a first preset height value and a second preset height value corresponding to the grid to form a collision zone, and only retains the point cloud with the height located within the collision zone as candidate points; The occupied grid generation module is used to mark grids with a number of candidate points reaching a preset threshold as obstacle grids, and otherwise mark them as passable grids or grids with no data, thereby generating an occupied grid map.

[0016] Thirdly, this application provides a computer device, including: one or more processors, and a memory; The memory stores computer-readable instructions that, when executed by the one or more processors, perform the steps of the method as described in the first aspect.

[0017] Fourthly, this application provides a storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the method described in the first aspect.

[0018] The technical effects of the raster map generation device based on point cloud slicing provided in the second aspect, the computer device provided in the third aspect, and the storage medium provided in the fourth aspect are described in the relevant description of the raster map generation method based on point cloud slicing provided in the first aspect. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the main process of the raster map generation method based on point cloud slicing in this application embodiment; Figure 2 This is a schematic diagram of the structure of the raster map generation device based on point cloud slicing according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation

[0020] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0021] The embodiments of this application are applied to raster map generation scenarios, and are particularly suitable for complex indoor environments with interference from slopes and ceilings. Existing technologies suffer from insufficient accuracy and reliability.

[0022] Therefore, in various embodiments of this application, 3D point cloud data of the target scene is acquired, mapped to a grid according to horizontal plane coordinates, and the lowest height value of the point cloud in each grid is taken as its local ground height. A first preset height value and a second preset height value corresponding to the grid are offset above the local ground height of each grid to form a collision zone. Only point clouds whose height is within the collision zone are retained as candidate points. For grids where the number of candidate points reaches a preset threshold, they are marked as obstacle grids; otherwise, they are marked as passable grids or grids with no data, generating an occupied grid map. Thus, this application achieves high-precision and high-reliability automated generation from point clouds to grid maps.

[0023] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0024] This application provides a method for generating raster maps based on point cloud slices, such as... Figure 1 As shown, the method includes: S1. Obtain the 3D point cloud data of the target scene, map it to a grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height.

[0025] In this embodiment, the 3D point cloud data is collected by LiDAR scanning and contains the 3D spatial coordinate information of each point in the target scene. The horizontal plane coordinates are the X-axis and Y-axis coordinates in the 3D point cloud data, and the height value is the Z-axis coordinate. The resolution of the grid can be set according to the actual scene requirements, for example, 0.1 meters to 0.5 meters. The higher the resolution, the higher the map accuracy, but the computational load also increases accordingly. Each grid may contain multiple point clouds. Since the ground is usually the lowest accessible surface within the grid range, the lowest Z value best represents the true ground position at that location. Therefore, the height value of the point with the smallest Z value is taken as the local ground height of the grid.

[0026] In this embodiment, after obtaining the 3D point cloud data of the target scene in step S1, the method further includes: The node tree of the point cloud tile index is recursively parsed, and the point cloud files corresponding to the leaf nodes without child nodes are extracted as the point cloud data to participate in the raster mapping.

[0027] The point cloud tile index is described in tileset.json format. In the node tree, parent nodes represent coarse sampled data, while leaf nodes represent the highest level of detail data. During recursive parsing of the node tree, it traverses layer by layer from the root node, extracting only the pnts format point cloud files corresponding to leaf nodes without child nodes. Combined with a memory-mapped caching mechanism, already parsed nodes are skipped when repeatedly generating raster maps, avoiding the computational overhead of repeated parsing. This ensures that the backend raster generation and frontend rendering use the same highest level of detail data, preventing the omission of ghost obstacles or real obstacles due to inconsistent data sources.

[0028] S2. Offset the first and second preset height values ​​corresponding to the grid above the local ground height of each grid to form a collision zone, and retain only the point cloud with the height within the collision zone as candidate points.

[0029] The collision zone is defined as a height range. The first preset height is the ground thickness preset value, which can be 0.3 meters, used to filter ground noise point clouds below the robot's chassis. The second preset height is the safe passage height preset value, which can be 1.5 meters, used to exclude ceiling point clouds and point clouds of objects above the robot's top. Each grid independently calculates its local ground height and forms a corresponding collision zone. Therefore, the upper and lower boundaries of the collision zones for different grids may differ. This grid-by-grid independent approach makes this method applicable to complex terrains such as ramps and steps. It should be noted that the above values ​​are only examples and can be adjusted according to the robot size and scene characteristics in actual applications. For example, for small robots, the first preset height value can be set to 0.1 meters; for large warehouse robots, the second preset height value can be set to 2.0 meters.

[0030] This embodiment uses a step-by-step floor detection combined with collision zone slicing to naturally exclude the ceiling from the floor. Compared to global planar fitting: a. The ground height difference between different floors can be 3~5m, and the global planar height value cannot be compatible with all floors.

[0031] b. The indoor floor has non-planar structures such as ramps, steps, and speed bumps. Global planar fitting has large residuals for these uneven areas, making it prone to misjudgment.

[0032] c. Ceiling height is usually 2.5~3.5m. The global fixed height tolerance cannot be compatible with all floors. If the value is set too high, low obstacles will be missed. If the value is set too low, the ceiling will still be misjudged.

[0033] For grid-by-grid floor inspection, each grid independently determines the ground height, naturally adapting to ramps and steps; collision zones are defined with relative offset based on the local ground, and ceilings naturally fall above the zone and are excluded, without needing to know the floor structure in advance.

[0034] In one example, in a multi-story warehouse scenario with two floors connected by ramps, the global plane fitting method generates a large number of false obstacle detections due to interference from the ceiling and ramps, while the step-by-step floor detection effectively eliminates such false detections, significantly improving the obstacle detection accuracy.

[0035] S3. For grids where the number of candidate points reaches a preset threshold, mark them as obstacle grids; otherwise, mark them as passable grids or grids with no data, and generate an occupied grid map.

[0036] Candidate points are point clouds located within the collision zone. The number of candidate points falling into the collision zone in each grid cell reflects the likelihood of an obstacle at that location: the more candidate points, the higher the probability of a real obstacle at that location. A preset threshold is used to determine the number of candidate points; grid cells with a candidate point count reaching the preset threshold are marked as obstacle cells, otherwise they are marked as passable cells or cells with no data.

[0037] In this embodiment, the process of generating the preset threshold in step S3 includes: The preset threshold AMOP (adaptive minimum obstacle points) is dynamically calculated based on the global average number of points per grid in the 3D point cloud data. AMOP = max(int(avg×a), b); In the formula, avg is the global average number of points per grid in the 3D point cloud data, int is the floor function, max is the maximum value function, a is the preset scaling factor, and b is the preset lower limit protection value.

[0038] The global average number of points per grid cell (avg) is calculated by simultaneously accumulating the total number of points and the total number of grid cells where points fall during a single scan in step S1, and then dividing the two to obtain avg. In one example, 'a' is 0.05, which is 5% of the average number of points, and its value ranges from 0.01 to 0.2 to accommodate different point cloud densities. 'b' is 10, ensuring that even in extremely sparse point clouds, very few noise points are not identified as obstacles, and its value ranges from 6 to 20.

[0039] For each candidate obstacle cell, the number of candidate points n within the collision zone is counted. Only when n ≥ AMOP is the cell confirmed as an obstacle cell.

[0040] In one example, a sparse warehouse has 5 million point clouds and 120,000 rasters, so the global average number of points per cell (avg) is 42, and the preset threshold AMOP is 10. A dense office building has 30 million point clouds and 250,000 rasters, so the global average number of points per cell (avg) is 120, and the preset threshold AMOP is 10. An ultra-dense substation has 80 million point clouds and 300,000 rasters, so the global average number of points per cell (avg) is 267, and the preset threshold AMOP is 13.

[0041] Therefore, the core value of the adaptive threshold in this embodiment lies in extreme scenarios with uneven point cloud density distribution: when there are occlusions or scanning blind spots in a localized area, the global average number of points per grid cell (avg) may differ significantly in different regions, and a fixed threshold cannot cover all areas. The adaptive threshold enables each scene to obtain a reasonable judgment benchmark that matches its point cloud density.

[0042] After marking obstacle grids, each grid is ternarily encoded: grids with infinite local ground height (i.e., no point cloud data) are encoded as 0, indicating a no-data state; grids with confirmed obstacles are encoded as 2, indicating an obstacle state; and other grids with ground points but not confirmed as obstacles are encoded as 1, indicating a passable state. This ternary encoding enables the robot to clearly distinguish between the three states: no point cloud data, passable, and obstacle, effectively preventing the robot from traversing unscanned areas.

[0043] In one embodiment, after generating the occupancy raster map, the method further includes: Morphological opening is performed on the occupied raster map. The erosion rule is executed first, followed by the dilation rule. The erosion rule is that if the number of obstacle cells in the eight-neighborhood of an obstacle cell is less than a preset degradation threshold, the obstacle cell is downgraded to a passable cell. The dilation rule is that if the number of obstacle cells in the eight-neighborhood of a passable cell reaches a preset recovery threshold, the passable cell is restored to an obstacle cell. The calculated occupied grid map is filtered by connected component area. For obstacle connected components with physical areas smaller than a preset area threshold, they are filtered and removed based on physical area, point cloud density, height, and neighborhood continuity. The remaining obstacle connected components are retained as real obstacle areas in the occupied grid map.

[0044] In this embodiment, the dilation in the morphological opening operation is as follows: the preset degradation threshold is 2, so if there are fewer than 2 obstacle neighbors in the eight-neighborhood, the grid is downgraded to ground, which is a passable grid; the preset recovery threshold is 5, so if a passable grid has 5 or more obstacle neighbors, it is restored to an obstacle grid. Thus, noise from isolated 1-2 grids is eroded and not recovered by dilation; linear obstacles of 1-2 grid width are eroded and then recovered by dilation due to their high neighbor density, such as thin walls.

[0045] The parameters are as follows: Area refers to the projected physical area of ​​the connected component on a horizontal plane, used to filter out isolated noise points that are too small; Point Cloud Density refers to the number of candidate points falling into the connected component per unit area, used to distinguish between sparse noise and dense obstacles; Height refers to the difference between the maximum and minimum height of the point cloud within the connected component, used to exclude low-lying ground noise; Neighborhood Continuity refers to the ratio of obstacle grids to the total number of grids within a perimeter of the connected component, used to identify real obstacles connected to surrounding structures. The connected component is identified as isolated noise and removed only when the area, point cloud density, height span, and neighborhood continuity are all less than their respective preset thresholds; otherwise, it is retained as a real obstacle area, thus ensuring that small real obstacles such as thin pillars, equipment bases, and road bollards are not mistakenly deleted while filtering out noise.

[0046] The system employs a three-layer noise filtering approach, which involves a first layer of filtering based on the number of collision points, a second layer of filtering based on morphological opening operations, and finally, filtering based on the area of ​​connected components.

[0047] It should be noted that the three layers of noise filtering are interdependent: First-layer filtering: Based on point density, it filters isolated barrier cells caused by sparse noise, relying on an adaptive threshold.

[0048] Second layer filtering: relies on the dense noise already filtered by the first layer.

[0049] The third layer of filtering is performed after the second layer of filtering. After the isolated noise is greatly compressed, the area threshold can more accurately separate noise clusters from real small obstacles.

[0050] Therefore, from the first layer of filtering to the second layer, the residual noise mainly manifests as scattered 1-2 isolated obstacles, which is exactly the mode that morphological erosion in the second layer of filtering is best at handling. From the second layer of filtering to the third layer of filtering, the noise in the second layer of filtering is compressed into smaller connected components, and the preset area threshold can more accurately separate noise clusters from real small obstacles, such as column bases, which are usually >0.5m².

[0051] In this embodiment, after generating the occupied grid map in step S3, the method further includes: Obtain manually annotated semantic polygons; For semantic polygons of type wall, set the grid within the coverage area of ​​the semantic polygon to be impassable; For semantic polygons of type risk region, the passage cost of the raster within the coverage area of ​​the semantic polygon is superimposed with the semantic risk weight, and the semantic risk weight is used to influence path search; For impassable graticles, perform distance cost inflation, applying a decay cost to surrounding graticles based on Chebyshev distance: cost = c × (1 - (cheb-1) / R); R = ceil(d / r); In the formula, c is the preset attenuation coefficient, cheb is the Chebyshev distance from the current grid to the nearest impassable grid, R is the preset expansion radius, ceil is the round-up function, d is the preset safety distance constant, and r is the grid resolution.

[0052] In one example, the preset attenuation coefficient c is 5.0, and the safety distance constant d is 0.3. In other examples, the preset attenuation coefficient c ranges from 1.0 to 10.0, and the safety distance constant d ranges from 0.1 to 0.5 m.

[0053] The way semantic annotation and inflation cost are superimposed depends on the annotation type: (1) Wall type is marked as hard overlay: The scan line algorithm is used to directly set all grid cells in the area covered by the wall polygon as impassable. The subsequent Chebyshev dilation and heuristic search algorithm pathfinding are all regarded as impassable obstacles.

[0054] (2) Risk area type is labeled as soft overlay: Instead of hard modification to the raster encoding, the semantic risk weight is written into an independent soft cost map. When the heuristic search algorithm is navigating, the neighbor node passage cost is based on the basic movement cost plus the soft cost map value. The path can still traverse the risk area, but because the cost is much higher than that of ordinary passable grids, it naturally prefers to detour. The semantic risk weight can be set to 10.

[0055] (3) Superposition of Chebyshev inflation cost: The total travel cost during pathfinding in the heuristic search algorithm is a superposition of three layers: totalCost = basic movement cost + Chebyshev inflation cost + semantic risk cost. The three layers of cost are calculated independently and accumulated cell by cell.

[0056] In one example, combining the above steps and taking warehouse point cloud data as an example, the system first recursively parses the tile index and extracts the leaf node point cloud, performing a single scan at a resolution of 0.2 meters to construct a height map. Then, each grid cell is sliced ​​using collision zones ranging from 0.3 meters to 1.5 meters. Combining adaptive point count verification, ternary encoding, and a three-layer filtering mechanism, shelves and columns are encoded as obstacle cells, aisle floors as passable cells, and unscanned corners as cells with no data. This results in a high-precision and highly reliable raster map.

[0057] In one embodiment, when applying the occupancy grid map generated in the above embodiments to indoor navigation, the method further includes: S41. A heuristic search algorithm is used to search for paths on an occupied grid map with superimposed semantic risk weights. The heuristic function is the Octile distance h. h=max(|dx|,|dy|)+(√2-1)×min(|dx|,|dy|); In the formula, dx and dy are the grid distances from the current node to the target node in two orthogonal directions on the horizontal plane, respectively.

[0058] The heuristic search algorithm expands nodes in an eight-directional neighborhood, allowing movement in eight directions, including diagonal directions. During the search, the cost of each node is the sum of the basic movement cost, the semantic risk cost in the soft-cost graph, and the Chebyshev inflation cost. The Octile distance heuristic function matches the eight-directional movement model, ensuring high search efficiency while finding the optimal path.

[0059] S42. Perform line-of-sight optimization on the searched path. Use the Bresenham line detection algorithm to detect the straightness between path nodes. If there is no grid with a cost greater than 0 between two nodes, delete the intermediate redundant node.

[0060] S43. Perform Douglas-Peucker simplification on the path after the line-of-sight shortcut is processed, output the simplified navigation waypoint sequence, and calculate the heading angle of each waypoint. The heading angle of the intermediate waypoint is taken from the tangent direction of the line connecting the preceding and following waypoints. The heading angles of the first and last waypoints can be specified by the user or taken from the direction of the adjacent waypoint.

[0061] The core of the line-of-sight optimization lies in using the Bresenham line detection algorithm to determine whether there are obstacles cutting between the first and last nodes of a path when there are three or more consecutive nodes. If there are no grids with a cost greater than 0 on the straight path between two nodes, then these two nodes can pass directly, and all intermediate nodes are redundant and are deleted. This step can significantly reduce the number of nodes in the path while maintaining the actual feasibility of the path.

[0062] The Douglas-Peucker simplified tolerance parameter is preferably 1.0 meter, meaning that intermediate nodes deviating from the line connecting the beginning and end of the path by less than 1.0 meter are removed while maintaining the overall shape of the path. The heading angle is calculated as follows: for intermediate waypoints, the direction of the displacement vector from the preceding waypoint to the following waypoint is taken as the heading angle; for the first waypoint, the starting direction is used or specified by the user; for the last waypoint, the direction from the second-to-last waypoint to the last waypoint is taken or specified by the user.

[0063] In this embodiment, after the Douglas-Peucker simplification is completed, a secondary collision check is performed on the connections between all adjacent waypoints in the simplified navigation waypoint sequence. Specifically, the grids traversed by each adjacent waypoint connection are checked sequentially to determine if there are any obstacle grids, unknown grids, or risk areas. Simultaneously, the distance between the connection and the nearest obstacle is checked to ensure it meets the robot's collision avoidance radius requirements. If all adjacent waypoint connections meet the safety requirements, the simplified waypoint sequence is output. If any connections do not meet the safety requirements, the node with the largest contribution from the deleted nodes is restored, the waypoint sequence is regenerated, and the check is performed again until all adjacent waypoint connections meet the safety requirements. This secondary check mechanism ensures that no new collision risks arise during the line-of-sight simplification and Douglas-Peucker simplification process due to node deletion, guaranteeing the navigation safety of the output waypoint sequence.

[0064] In summary, the above embodiments have the following effects: First, by independently calculating the local ground height in each grid and offsetting it to form a collision zone, the ceiling point cloud naturally falls above the collision zone and is excluded, while the ground noise point cloud falls below the collision zone and is filtered out. This fundamentally eliminates the interference of the indoor ceiling on obstacle detection, which is different from the fixed height threshold or global plane fitting method of the existing technology.

[0065] Second, by using a three-value encoding method to mark grids with a preset threshold of candidate points as obstacle grids, otherwise as passable grids or grids with no data, or to retain the state of no data, the grid map can clearly distinguish between three states: point cloud and safe, point cloud and obstacle, and no point cloud data, thus avoiding mislabeling unscanned areas as passable areas.

[0066] Third, by using an adaptive point count threshold to determine the number of candidate points in each grid, the obstacle confirmation threshold is dynamically adjusted according to the density of the scene point cloud, avoiding the problem of missed detections in sparse areas and false detections in dense areas with a fixed threshold; combined with a three-layer noise filtering mechanism of morphological opening operation and connected component area filtering, isolated noise is filtered out while real structures such as thin walls are preserved, and the grid quality is significantly improved.

[0067] Fourth, by recursively parsing the tile index and extracting only the leaf node point cloud, it ensures that the backend raster generation and the frontend rendering use the same highest level of detail data, avoiding the omission of ghost obstacles and real obstacles caused by coarse sampling of the parent layer.

[0068] In summary, this application starts with point cloud data and achieves the automated generation of high-precision, high-reliability occupancy grid maps from 3D point clouds through collision zone slicing, adaptive threshold determination, ternary encoding, and three-layer noise filtering.

[0069] In one embodiment, such as Figure 2 As shown, this application also provides a raster map generation apparatus based on point cloud slices, including: The point cloud acquisition module 201 is used to acquire the three-dimensional point cloud data of the target scene, map it to the grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height. The collision zone construction module 202 is used to offset the first preset height value and the second preset height value corresponding to the grid above the local ground height of each grid to form a collision zone, and only retain the point cloud with the height located within the collision zone as candidate points; The occupied grid generation module 203 is used to mark grids with the number of candidate points reaching a preset threshold as obstacle grids, and otherwise mark them as passable grids or grids with no data, thereby generating an occupied grid map.

[0070] In one embodiment, such as Figure 3 As shown, this application also provides a computer device 300, comprising: The communication interface 301 allows for information exchange with other devices or network nodes.

[0071] One or more processors 302 are connected to a communication interface 301 to enable information interaction with other devices or network nodes, and to execute the methods provided by one or more technical solutions in the above embodiments when running computer programs.

[0072] Memory 303 is used to store computer-readable instructions that can be executed on processor 302. When executed by one or more processors 302, the computer-readable instructions perform the steps of the raster map generation method based on point cloud slicing in the above embodiments.

[0073] The computer device 300 in this embodiment only shows a block diagram of a portion of the structure related to the present application solution, and does not constitute a limitation on the computer device to which the present application solution is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, such as power supply, input / output interfaces, etc. Furthermore, the computer device 300 of this embodiment can operate on an operating system stored in memory 303, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0074] In one embodiment, this application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the raster map generation method based on point cloud slicing in the above embodiments.

[0075] The computer-readable instructions in the aforementioned computer device 300 and storage medium may be application programs.

[0076] In addition, the specific descriptions of the technical effects and steps of the raster map generation apparatus based on point cloud slicing, the computer device 300, and the storage medium in the above embodiments are all based on the relevant descriptions of the embodiments of the raster map generation method based on point cloud slicing.

[0077] Since the systems / devices described in the above embodiments of this application are systems / devices used to implement the methods of the above embodiments of this application, those skilled in the art can understand the specific structure and modifications of the system / devices based on the methods described in the above embodiments of this application, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of this application fall within the scope of protection of this application.

[0078] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0080] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The words "a" or "an" preceding a component do not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0081] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0082] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0083] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, then this application should also include these modifications and variations.

Claims

1. A method for generating raster maps based on point cloud layer slicing, characterized in that, include: Acquire 3D point cloud data of the target scene, map it to a grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height; Offset the local ground height of each grid by a first preset height value and a second preset height value corresponding to the grid to form a collision zone, and retain only the point cloud with a height within the collision zone as candidate points; For grids where the number of candidate points reaches a preset threshold, they are marked as obstacle grids; otherwise, they are marked as passable grids or grids with no data, and an occupied grid map is generated.

2. The method according to claim 1, characterized in that, After acquiring the 3D point cloud data of the target scene, the process further includes: The node tree of the point cloud tile index is recursively parsed, and the point cloud files corresponding to the leaf nodes without child nodes are extracted as the point cloud data to participate in the raster mapping.

3. The method according to claim 1, characterized in that, The first preset height value is the ground thickness preset value; the second preset height value is the ceiling height preset value; the local ground height offset from the first preset height value is used as the lower boundary of the collision zone; and the local ground height offset from the second preset height value is used as the upper boundary of the collision zone.

4. The method according to claim 1, characterized in that, The process of generating the preset threshold includes: A preset threshold AMOP is dynamically calculated based on the global average number of points per grid in the 3D point cloud data. AMOP = max(int(avg×a), b); In the formula, avg is the global average number of points per grid in the three-dimensional point cloud data, int is the floor function, max is the maximum value function, a is the preset scaling factor, and b is the preset lower limit protection value.

5. The method according to claim 1, characterized in that, After generating the occupied raster map, the process also includes: Morphological opening is performed on the occupied grid map. First, the erosion rule is executed, and then the dilation rule is executed. The erosion rule is that if the number of obstacle cells in the eight-neighborhood of an obstacle cell is less than a preset degradation threshold, then the obstacle cell is downgraded to a passable cell. The dilation rule is that if the number of obstacle cells in the eight-neighborhood of a passable cell reaches a preset recovery threshold, then the passable cell is restored to an obstacle cell. The calculated occupied grid map is filtered by connected component area. For obstacle connected components with physical areas smaller than a preset area threshold, they are filtered and removed based on physical area, point cloud density, height, and neighborhood continuity. The remaining obstacle connected components are retained in the occupied grid map as real obstacle areas.

6. The method according to any one of claims 1 to 5, characterized in that, After generating the occupied raster map, the process also includes: Obtain manually annotated semantic polygons; For semantic polygons of type wall, set the grids within the coverage area of ​​the semantic polygon to be impassable; For semantic polygons of type risk region, the passage cost of the raster within the coverage area of ​​the semantic polygon is superimposed with semantic risk weight; For impassable graticles, perform distance cost inflation, and apply a decay cost to surrounding graticles based on the Chebyshev distance: cost = c × (1 - (cheb-1) / R); R = ceil(d / r); In the formula, c is the preset attenuation coefficient, cheb is the Chebyshev distance from the current grid to the nearest impassable grid, R is the preset expansion radius, ceil is the round-up function, d is the preset safety distance constant, and r is the grid resolution.

7. The method according to claim 6, characterized in that, After generating the occupied raster map, the process also includes: A heuristic search algorithm is used to search for paths on an occupancy raster map with semantic risk weights superimposed. The heuristic function is the Octile distance h. h=max(|dx|,|dy|)+(√2-1)×min(|dx|,|dy|); In the formula, dx and dy are the grid distances from the current node to the target node in two orthogonal directions on the horizontal plane, respectively; The searched path is streamlined by using the Bresenham line detection algorithm to check the straightness between path nodes. If there is no grid with a cost greater than 0 between two nodes, the intermediate redundant node is deleted. The path after the line-of-sight shortcut is simplified using Douglas-Peucker simplification, outputting a simplified navigation waypoint sequence and calculating the heading angle of each waypoint. The heading angle of the intermediate waypoint is taken from the tangent direction of the line connecting the preceding and following waypoints, while the heading angles of the first and last waypoints are specified by the user or taken from the direction of the adjacent waypoint.

8. A raster map generation device based on point cloud layer slicing, characterized in that, include: The point cloud acquisition module is used to acquire 3D point cloud data of the target scene, map it to a grid according to the horizontal plane coordinates, and take the lowest height value of the point cloud in each grid as its local ground height. The collision zone construction module is used to offset the local ground height of each grid by a first preset height value and a second preset height value corresponding to the grid to form a collision zone, and only retains the point cloud with the height located within the collision zone as candidate points; The occupied grid generation module is used to mark grids with a number of candidate points reaching a preset threshold as obstacle grids, and otherwise mark them as passable grids or grids with no data, thereby generating an occupied grid map.

9. A computer device, characterized in that, include: One or more processors, and memory; The memory stores computer-readable instructions that, when executed by the one or more processors, perform the steps of the method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the method as described in any one of claims 1 to 7.

Images