A method, system, and unmanned truck for identifying potholes on roads in mining areas

Abstract

This invention discloses a method, system, and unmanned truck for identifying potholes on roads in mining areas, within the field of road surface pothole recognition technology. The method includes: acquiring point cloud data of the road; downsampling the acquired point cloud data to obtain a first dataset; clustering the first dataset to obtain clustered point clouds; segmenting ground point clouds from the clustered point clouds; performing planar fitting on the ground point clouds to obtain a fitting plane; calculating the height value of the centroid of each clustered point cloud projected onto the fitting plane, and outputting ground point clouds with height values ​​less than a set value as potholes. This method can accurately detect potholes on roads, providing a basis for rationally planning driving routes.

Description

Technical Field

[0001] This invention belongs to the field of road surface pothole recognition technology, specifically relating to a method, system, and unmanned truck for recognizing road surface potholes in mining areas. Background Technology

[0002] With the continuous development of autonomous driving technology, more and more ore transport vehicles in the mining machinery field are adopting autonomous driving technology, which can effectively reduce labor costs and improve the production efficiency of mining operations. In structured roads, obstacle detection algorithms are numerous and relatively mature. However, in unstructured roads like those in mining areas, the heavy loads of mining trucks easily lead to complex road surfaces such as potholes and bulges, and the terrain is uneven. For the entire autonomous driving system, the condition of the road directly affects the smooth operation of ore transport vehicles. Therefore, it is necessary to rely on the autonomous driving perception system to accurately detect potholes in mining roads. If the ore transport vehicle enters a pothole, inaccurate detection results may cause the truck to violently shake, resulting in ore or soil spilling onto the road. This not only affects the normal driving of other vehicles and creates significant safety hazards, but also causes irreversible damage to the mechanical components of the ore transport vehicle itself. Timely and accurate detection of potholes allows the decision-making system to design a reasonable driving path for the ore transport vehicle, thereby improving the overall performance of the autonomous driving system. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method, system, and unmanned truck for identifying potholes on mining roads, which can accurately detect potholes on roads and provide a basis for rationally planning driving routes.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, a method for identifying road surface potholes in mining areas is provided, comprising: acquiring point cloud data of the road; downsampling the acquired point cloud data to obtain a first dataset; clustering the first dataset to obtain clustered point clouds; segmenting ground point clouds from the clustered point clouds; performing plane fitting on the ground point clouds to obtain a fitting plane; calculating the height value of the centroid of each clustered point cloud projected onto the fitting plane, and outputting ground point clouds with height values ​​less than a set value as potholes.

[0006] Furthermore, the downsampling method is as follows: a voxel grid method based on VoxelGrid filter is used for downsampling, and a three-dimensional grid is built from the input point cloud data. In each voxel grid, the centroid is used to represent other point clouds in the grid.

[0007] Further, the clustering specifically includes: performing coarse clustering on the point cloud data in the first dataset using Euclidean clustering; calculating the contour, centroid, and Euclidean distance between the centroid and the lidar of the coarse clustered point cloud; calculating the precision operation factor required for fine clustering based on the precision of the coarse clustering, the adjustment ratio, and the Euclidean distance between the centroid and the lidar; performing Euclidean fine clustering based on the calculated precision operation factor to obtain the fine clustered point cloud; traversing the fine clustered point cloud and updating the centroid coordinates and contour of the fine clustered point cloud.

[0008] Furthermore, the precision calculation factor is obtained through the following formula:

[0009]

[0010] Where W represents the precision operation factor, D represents the coarse clustering precision, R represents the clustering precision adjustment ratio, and L represents the coarse clustering distance.

[0011] Furthermore, the step of segmenting the ground point cloud from the clustered point cloud specifically involves: applying a cropBox filter to perform pass-through filtering on all clustered point clouds, treating the clustered point clouds within the bounding box of the cropBox filter as ground point clouds, and treating the clustered point clouds outside the bounding box as non-ground point clouds.

[0012] Furthermore, all clustered point cloud data are traversed, invalid empty point cloud data are filtered out, and then the ground point cloud is filtered out by searching the k nearest neighbor clustered point cloud data through the establishment of KD-tree.

[0013] Furthermore, the least squares method is used to perform plane fitting on the ground point cloud to obtain the fitted plane.

[0014] Secondly, a system for identifying road surface potholes in mining areas is provided, comprising: a data acquisition module for acquiring point cloud data of the road; a first data processing module for downsampling the acquired point cloud data to obtain a first dataset; a second data processing module for clustering the first dataset to obtain a clustered point cloud; a third data processing module for segmenting ground point clouds from the clustered point clouds; a fourth data processing module for performing plane fitting on the ground point clouds to obtain a fitting plane; and a judgment module for calculating the height value of the centroid projection of each clustered point cloud onto the fitting plane, and outputting ground point clouds with height values ​​less than a set value as potholes.

[0015] Thirdly, a system for identifying road surface potholes in mining areas is provided, comprising a processor and a storage device, wherein the storage device stores multiple instructions for the processor to load and execute the steps of the method described in the first aspect.

[0016] Fourthly, an unmanned truck for transporting ore is provided, the truck being equipped with the mine road surface pothole recognition system described in the second aspect.

[0017] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0018] (1) This invention acquires point cloud data of roads in complex road conditions in mining areas, clusters the acquired point cloud data, segments the ground point cloud, performs plane fitting on the ground point cloud to obtain the fitting plane, calculates the height value of the centroid of each ground point cloud projected onto the fitting plane, and outputs ground point clouds with height values ​​less than the set value as pits; it can accurately detect pits on the road and provide a basis for rationally planning driving routes.

[0019] (2) The present invention adopts a combination of coarse clustering and fine clustering to solve the problem that the accuracy of relying solely on coarse clustering as a single segmentation method is not high, effectively improving the point cloud clustering effect for complex mining environments, which is beneficial to the next step of point cloud segmentation;

[0020] (3) The present invention uses the cropBox filtering method, which is different from the commonly used voxel grid method. For the point cloud clustering of polygons, it performs through filtering in the X, Y and Z axis directions, which can effectively filter out point clouds according to the actual terrain and is suitable for complex terrain scenes.

[0021] (4) This invention uses least squares plane fitting and cluster centroid to detect road potholes and makes a judgment based on the absolute height difference, which has better robustness.

[0022] (5) Based on the identification of road surface potholes in this invention, the driving path of the unmanned truck is reasonably planned, the driving process of the unmanned truck is more stable, the damage to the unmanned truck is less, and the impact on the environment is less. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the main process of a method for identifying road surface potholes in mining areas provided by an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0025] Example 1:

[0026] A method for identifying road surface potholes in mining areas includes: acquiring point cloud data of the road; downsampling the acquired point cloud data to obtain a first dataset; clustering the first dataset to obtain clustered point clouds; segmenting ground point clouds from the clustered point clouds; performing planar fitting on the ground point clouds to obtain a fitting plane; calculating the height value of the centroid of each clustered point cloud projected onto the fitting plane, and outputting ground point clouds with height values ​​less than a set value as potholes.

[0027] like Figure 1 As shown, the method for identifying road potholes specifically includes the following steps:

[0028] Acquire point cloud data of the road; downsample the acquired point cloud data to obtain the first dataset.

[0029] S1: The point cloud data of the mining area road acquired by the vehicle radar is downsampled based on the VoxelGrid voxel filter. The downsampling method is as follows: the voxel grid method based on the VoxelGrid voxel filter is used to downsample the input point cloud data to build a three-dimensional grid. In each voxel grid, the centroid is used to represent other point clouds in the grid. The point cloud processed in this way can reduce the point cloud data, thereby improving the computing speed.

[0030] Clustering is performed on the first dataset to obtain a clustered point cloud; the clustering methods include:

[0031] S2: Euclidean clustering is used to perform coarse clustering on the point cloud data in the first dataset. For a point P in space, the KD-Tree nearest neighbor search algorithm is used to find k points closest to P. Points whose distance to P is less than a set threshold are clustered into set Q. If the number of elements in Q no longer increases, the entire clustering process ends; otherwise, points other than P must be selected from set Q, and the above process is repeated until the number of elements in Q no longer increases. In this embodiment, the KD-Tree nearest neighbor search algorithm traverses the point cloud data in the first dataset, and points within a certain range whose distance is less than a threshold are considered to be a point cloud cluster, until the number of points in that cluster no longer increases.

[0032] S3: Calculate the outline, centroid, and Euclidean distance between the centroid and the LiDAR of the coarse clustered point cloud; iterate through all point cloud data, and average the maximum and minimum values ​​of the coarse clustered point cloud on the three coordinate axes to obtain the coordinates of the centroid of the coarse clustered point cloud; calculate the relative distance between the centroid of each coarse clustered point cloud and the LiDAR in the vehicle coordinate system. The Euclidean distance between the centroid and the LiDAR is obtained using the following formula:

[0033]

[0034] Where d represents the Euclidean distance between the centroid and the lidar, and x i The x-coordinate of the i-th cluster centroid is represented by y. j Let x represent the ordinate of the i-th cluster centroid, x represent the abscissa of the lidar, and y represent the ordinate of the lidar.

[0035] S4: Calculate the accuracy factor required for fine clustering based on the accuracy of coarse clustering, the adjustment ratio, and the Euclidean distance between the centroid and the lidar. The accuracy factor required for fine clustering is jointly determined by the coarse clustering accuracy parameters, the clustering accuracy adjustment ratio, and the coarse clustering distance. The accuracy factor is obtained using the following formula:

[0036]

[0037] Where W represents the precision operation factor, D represents the coarse clustering precision, R represents the clustering precision adjustment ratio, and L represents the coarse clustering distance.

[0038] S5: Perform Euclidean fine clustering based on the required precision operation factor to obtain the fine clustered point cloud.

[0039] S6: Traverse the fine cluster point cloud, update the centroid coordinates and contour of the fine cluster point cloud, and finally obtain the cluster point cloud.

[0040] Segmenting ground point clouds from clustered point clouds.

[0041] S7: Segment the ground point cloud from the clustered point cloud. Specifically, use a cropBox filter to perform pass-through filtering on all clustered point clouds. The clustered point clouds within the bounding box of the cropBox filter are considered as ground point clouds, and the clustered point clouds outside the bounding box are considered as non-ground point clouds.

[0042] S8: The KD-Tree point cloud search algorithm is adopted to accelerate the point cloud search speed. It traverses all clustered point cloud data, filters out invalid empty point cloud data, and then searches the k nearest neighbor clustered point cloud data by building a KD-tree to filter out the ground point cloud, which can improve the calculation speed of point cloud.

[0043] Plane fitting is performed on the ground point cloud to obtain the fitted plane.

[0044] S9: The surface normal vector of the point cloud data is extracted using a plane fitting method. Simultaneously, by optimizing the minimum plane error, the complexity of the plane fitting calculation is simplified. Then, a tangent plane under the least squares method is fitted onto the topological structure. In this embodiment, the least squares method is used to perform plane fitting on the ground point cloud to obtain the fitted plane. All clustered point cloud data are traversed, and the surface normal vector of the point cloud data is calculated according to the plane fitting equation, thereby determining the slope value relative to the xoy plane (the plane fitted by the ground point cloud).

[0045] Calculate the height value of the centroid projection of each cluster of point clouds onto the fitted plane, and output the ground point clouds with height values ​​less than the set value as pits.

[0046] S10: Calculate the height Z of the centroid projection of the clustered point cloud onto the fitting plane; with the rear axle center of the vehicle as the plane, calculate the average value of the mass points of the clustered point cloud on the nearby ground.

[0047] S11: Compare the average height of the centroids of the clustered point cloud with the set threshold. If it is less than the threshold, output the pitted point cloud data.

[0048] This embodiment acquires point cloud data of roads in complex mining areas, clusters the acquired point cloud data to segment ground point clouds, performs plane fitting on the ground point clouds to obtain a fitted plane, calculates the height value of the centroid of each ground point cloud projected onto the fitted plane, and outputs ground point clouds with height values ​​less than a set value as pits. This accurately detects pits on roads, providing a basis for rationally planning driving routes. This embodiment uses a combination of coarse and fine clustering to solve the problem of low accuracy when relying solely on coarse clustering, effectively improving the point cloud clustering effect for complex mining environments, which is beneficial for subsequent point cloud segmentation. This embodiment uses the cropBox filtering method, which, unlike the commonly used voxel raster method, performs direct filtering in the X, Y, and Z axes for polygonal point cloud clustering, effectively filtering out point clouds based on actual terrain, adapting to complex terrain scenes. This embodiment uses least squares plane fitting and cluster centroids to detect road pits, using absolute height difference for discrimination, which has better robustness.

[0049] Example 2:

[0050] Based on the method for identifying potholes on mining roads described in Embodiment 1, this embodiment provides a system for identifying potholes on mining roads, comprising: a data acquisition module for acquiring point cloud data of the road; a first data processing module for downsampling the acquired point cloud data to obtain a first dataset; a second data processing module for clustering the first dataset to obtain a clustered point cloud; a third data processing module for segmenting ground point clouds from the clustered point clouds; a fourth data processing module for performing planar fitting on the ground point clouds to obtain a fitting plane; and a judgment module for calculating the height value of the centroid of each clustered point cloud projected onto the fitting plane, and outputting ground point clouds with height values ​​less than a set value as potholes.

[0051] Example 3:

[0052] Based on the method for identifying road surface potholes in mining areas described in Embodiment 1, this embodiment provides a system for identifying road surface potholes in mining areas, including a processor and a storage device. The storage device stores multiple instructions for the processor to load and execute the steps of the method described in Embodiment 1.

[0053] Example 4:

[0054] Based on the pothole identification system for mining roads described in Embodiments 2 and 3, this embodiment provides an unmanned truck for transporting ore, wherein the truck is equipped with the pothole identification system for mining roads described in Embodiments 2 and 3.

[0055] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying road surface potholes in mining areas, characterized in that, include: Acquire point cloud data of the road; The acquired point cloud data is downsampled to obtain the first dataset; Cluster the first dataset to obtain a clustered point cloud; Segmenting ground point cloud from clustered point cloud; Perform planar fitting on the ground point cloud to obtain the fitted plane; Calculate the height value of the centroid projection of each cluster of point clouds onto the fitting plane, and output the ground point clouds with height values ​​less than the set value as pits; The clustering specifically includes: Euclidean clustering was used to perform coarse clustering on the point cloud data in the first dataset. The outline, centroid, and Euclidean distance between the centroid and the lidar of the coarse clustered point cloud were calculated. The accuracy calculation factor required for fine clustering is calculated based on the accuracy of coarse clustering, the adjustment ratio, and the Euclidean distance between the centroid and the lidar. Perform Euclidean fine clustering based on the required precision operation factor to obtain the fine clustered point cloud; Traverse the refined cluster point cloud and update the centroid coordinates and contour of the refined cluster point cloud.

2. The method for identifying road surface potholes in mining areas according to claim 1, characterized in that, The downsampling method is as follows: a voxel grid method based on VoxelGrid filter is used for downsampling. A three-dimensional grid is built from the input point cloud data. In each voxel grid, the centroid is used to represent other point clouds in the grid.

3. The method for identifying road surface potholes in mining areas according to claim 1, characterized in that, The precision calculation factor is obtained using the following formula: （2） Where W represents the precision calculation factor, D represents the coarse clustering precision, R represents the clustering precision adjustment ratio, and L represents the Euclidean distance between the centroid of the coarse clustered point cloud and the lidar.

4. The method for identifying road surface potholes in mining areas according to claim 1, characterized in that, The step of segmenting the ground point cloud from the clustered point cloud specifically involves: using a cropBox filter to perform pass-through filtering on all clustered point clouds, treating the clustered point clouds within the bounding box of the cropBox filter as ground point clouds, and treating the clustered point clouds outside the bounding box as non-ground point clouds.

5. The method for identifying road surface potholes in mining areas according to claim 1, characterized in that, Traverse all clustered point cloud data, filter out invalid empty point cloud data, and then search for the k nearest neighbor clustered point cloud data by building a KD-tree to filter out the ground point cloud.

6. The method for identifying road potholes in mining areas according to claim 1, characterized in that, The least squares method was used to fit the ground point cloud to a plane, and the fitted plane was obtained.

7. A system for identifying potholes on roads in mining areas, characterized in that, include: The data acquisition module is used to acquire point cloud data of the road. The first data processing module is used to downsample the acquired point cloud data to obtain the first dataset; The second data processing module is used to cluster the first dataset to obtain a clustered point cloud. The third data processing module is used to segment the ground point cloud from the clustered point cloud; The fourth data processing module is used to perform planar fitting on the ground point cloud to obtain the fitting plane; The judgment module is used to calculate the height value of the centroid projection of each cluster point cloud onto the fitting plane, and output the ground point cloud with a height value less than the set value as a pit; The clustering specifically includes: Euclidean clustering was used to perform coarse clustering on the point cloud data in the first dataset. The outline, centroid, and Euclidean distance between the centroid and the lidar of the coarse clustered point cloud were calculated. The accuracy calculation factor required for fine clustering is calculated based on the accuracy of coarse clustering, the adjustment ratio, and the Euclidean distance between the centroid and the lidar. Perform Euclidean fine clustering based on the required precision operation factor to obtain the fine clustered point cloud; Traverse the refined cluster point cloud and update the centroid coordinates and contour of the refined cluster point cloud.

8. A system for identifying potholes on roads in mining areas, characterized in that, It includes a processor and a storage device, wherein the storage device stores a plurality of instructions for the processor to load and execute the steps of the method according to any one of claims 1 to 6.

9. An unmanned truck for transporting ore, characterized in that, The truck is equipped with the identification system for potholes on the mining road surface as described in claim 7.

Images