A construction site retaining wall automatic identification method based on a point cloud map

Abstract

The application belongs to the technical field of engineering construction, and discloses a construction site retaining wall automatic identification method based on a point cloud map, which comprises the following steps: obtaining point clouds of a bulldozer working area in a point cloud map, performing coordinate conversion and spatial gridding, and obtaining a grid map; dividing the grid map into multiple height sections according to the point cloud height, selecting the height section with the largest number of grids as a ground section, filtering out the ground point cloud, and retaining the point cloud above the ground point cloud; clustering the non-ground point cloud, calculating the linearity of each cluster by using principal component analysis, and screening out retaining wall point cloud according to the linearity; further, the method further comprises the steps of performing secondary height segmentation on the retaining wall point cloud, selecting the section with the largest point cloud density for linear fitting, and obtaining the linear equation of the retaining wall. The application can effectively eliminate ground interference from complex construction environment point clouds, accurately identify the retaining wall by using linear features, and meet the requirements of real-time performance and accuracy of the bulldozer engineering site.

Description

Technical Field

[0001] This invention belongs to the field of engineering construction technology, and more specifically, relates to an automatic identification method for retaining walls at construction sites based on point cloud maps. Background Technology

[0002] With the development of intelligent construction machinery, autonomous operation of heavy machinery such as excavators and bulldozers on construction sites has become a research hotspot. In complex construction environments, environmental perception is a prerequisite for autonomous operation. When bulldozing, bulldozers need to push soil behind retaining walls. Therefore, identifying the location of retaining walls on the construction site before autonomous operation is crucial for the bulldozer's path planning and autonomous operation. Currently, bulldozers typically use sensors such as LiDAR and IMU combined with SLAM algorithms to construct a 3D point cloud map of the construction site. Identifying retaining walls within this point cloud map is the key challenge of this technology.

[0003] However, construction sites are characterized by unstructured environments, complex terrain, and abundant dust and uneven surfaces. The raw point cloud maps constructed using SLAM often contain massive amounts of noise and ground data, lacking semantic information. During bulldozing operations, retaining walls are crucial reference points, but existing point cloud processing methods struggle to accurately and quickly extract their geometric features from the chaotic and disordered point cloud of the construction scene without significant manual intervention. Traditional line detection algorithms (such as Hough transform) are computationally intensive and susceptible to noise when directly applied to massive 3D point clouds, failing to meet the real-time and accuracy requirements of engineering sites. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides an automatic identification method for retaining walls in construction sites based on point cloud maps, the purpose of which is to improve the real-time performance and accuracy of automatic identification of retaining walls in construction sites.

[0005] To achieve the above objectives, this invention provides an automatic identification method for retaining walls at construction sites based on point cloud maps, comprising: Obtain the point cloud coordinates of the bulldozer working area in the point cloud map of the construction site, and then convert the point cloud coordinates to a standard coordinate system and perform spatial rasterization to obtain a raster map. The raster map is divided into multiple height segments according to the point cloud height. The height segment with the most grid cells is selected as the ground segment. Grid cells in the height segment whose point cloud height is greater than the point cloud height of the ground segment are selected as candidate grid cells. The angle between the average normal direction of the candidate grid cells and the ground is determined. Candidate grid cells with an angle less than a preset threshold are deleted. The point clouds in the remaining candidate grid cells are selected as non-ground point clouds. For the non-ground point cloud clusters, principal component analysis is used to calculate the linearity of each cluster, and several clusters with high linearity are selected as candidate clusters. Each candidate cluster is scored based on its boundary coverage rate, which is proportional to the score of the candidate cluster. The candidate cluster with the highest score is selected as the retaining wall point cloud. The boundary coverage rate is the ratio of the number of grid cells in the boundary region of the bulldozer working area in the candidate cluster to the total number of grid cells in the candidate cluster. Retaining wall identification is performed based on the point cloud of the retaining wall.

[0006] Furthermore, retaining wall identification is performed based on the retaining wall point cloud, including: The retaining wall point cloud is further segmented by height. The height segment with the most point cloud points is fitted with a straight line to obtain the straight line equation of the retaining wall. The straight line equation of the retaining wall is linearly interpolated to obtain the straight line of the retaining wall. The straight line of the retaining wall is then translated upward to fit the top line, thereby realizing the recognition of the retaining wall.

[0007] Further, the retaining wall is moved upwards in a straight line to fit the top line, including: Calculate the relative position of the centroid of the point cloud above the height segment with the most point cloud content to the centroid of the point cloud above the height segment with the most point cloud content, and then translate the retaining wall upwards in a straight line to fit the top line. Linear interpolation is performed on the straight line equation of the retaining wall, including: Calculate the difference between the maximum and minimum values ​​of the point cloud on each coordinate axis of the retaining wall straight line equation; select the coordinate axis with the larger difference as the main axis, and perform point interpolation along the main axis to obtain the retaining wall straight line.

[0008] Further, the average normal direction of the candidate grid is calculated, including: Traverse all points within each candidate grid, search for the k nearest neighbors of each point, perform principal component analysis on the k nearest neighbors to obtain the normal of the point; unify the orientation of the normal directions of all points within the candidate grid, and take the weighted average of the normals of each point and its k nearest neighbors as the normal vector of each point; add the normal vectors of all points within the candidate grid and normalize them to obtain the average normal direction of the candidate grid.

[0009] Furthermore, the candidate clustering score is calculated as follows:

[0010] in, The candidate cluster score is defined by edge_coverage, which represents the edge coverage of the candidate cluster, liners represents the principal scale length feature of the candidate cluster, sum_line represents the sum of the lengths of all candidate clusters, linearity represents the linearity of the candidate cluster, class_z represents the average height of the candidate cluster, and sum_z_average represents the average height of all candidate clusters.

[0011] Further, obtaining the raster map includes: Calculate the number of grid cells in the x and y axes in the standard coordinate system; Calculate the mean and variance of the point cloud height within each grid cell, and use the mean as the point cloud height of the corresponding grid cell and the variance as the variance of the corresponding grid cell. The average variance of the point cloud heights of all grid cells is calculated based on the point cloud height of each grid cell. If the difference between the variance of a grid cell and the average variance is greater than a preset threshold, the grid cell is removed as an outlier, and the grid map with outliers removed is obtained.

[0012] Furthermore, the bulldozer's working area is the expanded working area; obtaining the point cloud coordinates of the expanded working area includes: The expansion factor is calculated from the point cloud map based on the vertex coordinates of each vertex of the provided original working area. : ;in, The distance from each vertex of the original working region to the centroid of the original working region is given by expansion_distance, where expansion_distance is the set expansion distance. The expansion factor Multiply by the x and y coordinates of each vertex to obtain the coordinates of each vertex in the expanded working area; The point cloud map is subjected to intensity filtering and outlier filtering. Based on the coordinates of each vertex in the expanded working area and the filtered point cloud map, the point cloud coordinates within the expanded working area are obtained.

[0013] The present invention also provides an automatic identification system for retaining walls at construction sites based on point cloud maps, including a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the automatic identification method for retaining walls at construction sites based on point cloud maps described above.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the automatic identification method for retaining walls at construction sites based on point cloud maps as described in any of the preceding claims.

[0015] The present invention also provides a computer program product, including a computer program that, when the computer program is run on a computer, causes the computer to execute the automatic identification method for retaining walls at construction sites based on point cloud maps as described above.

[0016] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: (1) This invention provides a new method for automatic identification of retaining walls. Based on the histogram statistics of point cloud height, the ground point cloud is identified, the ground point cloud and the point cloud below its height are filtered out, the non-ground point cloud data above the ground are retained, and clustering and principal component analysis are combined to achieve fast and high-precision identification of retaining walls. Specifically, by segmenting the point cloud within the bulldozer's working area by height and identifying the ground segment, interference from ground point clouds and point clouds below their height is effectively filtered out, retaining non-ground point cloud data such as those above the ground, including retaining walls. Furthermore, the extracted non-ground point cloud data undergoes normal vector angle filtering to eliminate grids approximately parallel to the ground (grids with small angles), resulting in non-ground point cloud data areas that are significantly inclined or vertical to the ground. For noise on the ground, such as slopes or trees, the non-ground point clouds are clustered and subjected to principal component analysis. Considering that retaining walls are generally located at the boundaries of the working area, boundary coverage is used as an important indicator for scoring the clustered point clouds. The cluster with the highest score is considered the retaining wall point cloud. Thus, retaining wall point clouds are accurately and quickly identified. Moreover, this invention introduces principal component analysis into retaining wall recognition, which, compared to traditional line detection algorithms (such as Hough transform), has lower computational complexity and is more suitable for applications in massive 3D point cloud scenarios.

[0017] (2) Furthermore, since radar can only scan the surface of the retaining wall, and the top of the retaining wall is uneven and the straight line feature is not good, and the bottom is in too much contact with the ground point cloud, it is easy to be affected when performing straight line fitting on the retaining wall point cloud. Therefore, this invention further improves the accuracy of retaining wall recognition by dividing the retaining wall point cloud into height segments again. The point cloud height segment with the highest density has the highest linearity. The point cloud of this height segment is fitted with a straight line, and the straight line of the retaining wall is obtained by linear interpolation. The straight line is then translated to the upper point cloud to fit the top line.

[0018] (3) As a preferred option, when performing linear interpolation on the straight line equation of the retaining wall, selecting the axis with a longer span as the principal axis can ensure that the interpolation points are evenly distributed.

[0019] (4) Further, during the process of obtaining the raster map, the average variance of the height of all raster grid point clouds is used to characterize the average roughness of the entire working area. Based on this average variance, threshold filtering is performed on each grid to filter out abnormal grids. Only relatively flat areas such as ground and walls are retained in the obtained raster map, which is convenient for subsequent height layer analysis of the raster map.

[0020] (5) As a preferred option, the robustness of the method can be improved by expanding the working area of ​​the bulldozer.

[0021] Overall, this invention expands the working area to improve the robustness of the algorithm. By segmenting the point cloud map by height, it can effectively remove ground interference from the complex point cloud of the construction environment. It also uses linear features to accurately identify retaining walls, providing accurate environmental semantic information for the automatic operation of construction machinery such as bulldozers. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the automatic identification method for retaining walls at construction sites based on point cloud maps in an embodiment of the present invention. Figure 2 A flowchart of an automatic identification method for retaining walls at construction sites based on point cloud maps, provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the work area expansion provided in an embodiment of the present invention; Figure 4(a) is a schematic diagram of point cloud extraction of the working area provided in an embodiment of the present invention; Figure 4(b) is a schematic diagram of the point cloud within the expanded working area provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the point cloud meshing of the working area provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of non-ground point cloud data extracted after point cloud height segmentation, provided in an embodiment of the present invention. Figure 7 This is a diagram showing the operation results of clustering analysis on non-ground point clouds provided in an embodiment of the present invention; Figure 8 This is a diagram showing the point cloud recognition and extraction results of a retaining wall provided in an embodiment of the present invention; Figure 9 This is a diagram showing the result of further height segmentation of the point cloud of the retaining wall provided in an embodiment of the present invention; Figure 10 The final retaining wall straight line result diagram after fitting and interpolating the straight line provided in the embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0024] Example 1 This invention provides an automatic identification method for retaining walls at construction sites based on point cloud maps, such as... Figure 1 and Figure 2 As shown, it includes: S1, at the set expansion distance, expand the working area provided in the point cloud map to obtain the expanded working area in the point cloud map, and obtain the three-dimensional point cloud coordinates in the expanded working area.

[0025] In this embodiment of the invention, a three-dimensional point cloud map constructed by the bulldozer at the construction site using the SLAM algorithm is obtained. The expansion factor is calculated from the point cloud map based on the coordinates of the four vertices of the provided work area. The calculation process is as follows: Let the expansion distance be defined as `expansion_distance`, and the distance from each vertex of the original working region to the centroid of the working region be `distance`. The value of the expansion factor `factor` is: ; After obtaining the expansion factor, multiply it by the x and y values ​​of the four vertices to obtain the vertex coordinates of the expanded working region. The expanded working region is as follows: Figure 3 As shown, the four points are the four vertices of the original working area, and (X, Y, Z) in the figure are the vertex coordinates of the expanded working area. Simultaneously, intensity filtering and outlier filtering are performed on the point cloud map. In this embodiment, the SOR algorithm is used for outlier removal. The distances from the K nearest neighbors to each point in the point cloud map are calculated, and the average distance is calculated. The average distances of all points form a distribution, and points with larger average distances (closer to the distribution edge) are removed as outliers. Based on the vertex coordinates of the expanded working area and the filtered point cloud map, the point cloud coordinates within the expanded working area are extracted, as shown in Figures 4(a) and 4(b). Expanding the working area improves the robustness of the method.

[0026] S2 standardizes the point cloud coordinate system, transforming it to the standard coordinate system. That is, the point cloud coordinates in the expanded working area obtained in S1 are transformed from the point cloud coordinate system to the standard coordinate system, resulting in point cloud coordinates in the standard coordinate system.

[0027] In this embodiment of the invention, by translating the centroid coordinates of the expanded working area to the origin and rotating the entire point cloud to make the ground parallel to the xy plane in the standard coordinate system, the specific steps include: S21. In the point cloud coordinate system, calculate the centroid coordinates of the point cloud within the expanded working area, define the translation matrix, and set the value of the translation matrix to the centroid coordinates.

[0028] S22. Using the RANSAC ground fitting algorithm, the maximum allowable angle between the normal vector and the z-axis in the standard coordinate system is set to 15 degrees. The normal vector of the current ground is defined as the result of the fitting plane, and the target normal vector (0,0,1) is defined. Based on the values ​​of the two vectors, the rotation matrix is ​​determined. The rotation matrix and translation matrix are combined to obtain the transformation matrix. The point cloud of the working area is transformed to the desired standard coordinate system according to the transformation matrix.

[0029] In this embodiment of the invention, the centroid coordinates of the point cloud of the working area are set to [cx, cy, cz]. Then the translation matrix for:

[0030] The rotation matrix is ​​calculated by determining the angle between the fitting plane normal vector and the standard normal vector (0, 0, 1). :

[0031] The transformation matrix T is obtained by adding the rotation matrix and the translation matrix. All points in the working area are transformed to the standard coordinate system using the transformation matrix.

[0032] S3, Spatial Meshization, discretizes the continuous space composed of point cloud coordinates in the standard coordinate system into a regular network, and obtains the mean and variance of the z-value (height of the point cloud) of each grid cell. The average normal direction of each grid cell is calculated. Regions with large differences in z-values ​​are filtered out, retaining flat areas such as walls and the ground.

[0033] Specifically, it includes: S31. Determine the boundary of the point cloud in the standard coordinate system, calculate the minimum and maximum values ​​of the point cloud along the x and y axes, obtain the distance difference in the x and y directions (maximum value minus minimum value in the corresponding direction), and divide the point cloud into a grid according to the set grid size. Divide the distance difference by the set grid size to obtain the number of grid cells in the x and y directions. Traverse each point in the point cloud, assign it to the corresponding grid cell according to its coordinate value, and store its grid coordinates. A schematic diagram of the meshing process is shown below. Figure 5 As shown.

[0034] S32, iterate through all points within each grid cell, and calculate the mean and variance of the z-values ​​(representing the height information of the point cloud) of all points within that grid cell. The mean of the z-values ​​of all points within the grid cell is taken as the z-value of that grid cell, and the variance of the z-values ​​of all points within the grid cell is taken as the variance of that grid cell.

[0035] S33. Traverse each grid cell, iterating through all points within the cell. For each point, perform a k-nearest neighbor search and principal component analysis (PCA) on its k-nearest neighbors. These neighbors form a local surface. After PCA, the normal direction and curvature of this surface are obtained and used as the normal direction of the point. Unify the orientation of all points within the grid cell; in this embodiment, the orientation is unified to point towards the outer surface of the object. Calculate the weighted average of the normals of each point and its nearest neighbors to obtain the normal vector of each point, ensuring greater consistency in the direction of neighboring normals, thus obtaining the normal direction of all points within each grid cell. For each grid cell, sum the normal vectors of all points within it and then normalize them to obtain the average normal direction of the grid cell.

[0036] S34: Based on the z-value of each grid cell calculated in S2, calculate the average variance of the z-values ​​of all non-empty grid cells, and use it as a measure of the average roughness of the entire scene (working area). Iterate through all non-empty grid cells and perform a threshold filtering operation: if the difference between the variance value of a grid cell and the average variance is greater than a preset threshold, then the grid cell is considered an outlier and filtered out from the point cloud within the working area to obtain a grid map. In this embodiment of the invention, if the variance value of a grid cell is greater than twice the average variance, it is considered an outlier and filtered out to extract flat areas and retain relatively flat surfaces such as the ground and walls.

[0037] S4. Perform height layering analysis on the raster map to obtain the average height of the highest and lowest graticule in the raster map. Calculate the height difference between the two average heights and divide the height difference by the preset height resolution seg_h to obtain the total number of height segments (height segment index) for height division of the raster map. This achieves height layering of all graticules in the raster map, i.e., segmentation based on the z-value of the graticule. In this embodiment, the preset height resolution seg_h reference value is 0.15~0.2m. All graticules in the raster map are assigned to corresponding height segments, and the height segment with the most graticules is taken as the ground segment. Among all height segments, the point cloud of the height segment above the ground height (i.e., the height segment with a z-value greater than the ground segment) is taken for subsequent processing. Using the normal direction of all graticules calculated in S3, the angle (z-axis upward) between all graticules above the ground and the ground is judged. Gradually exclude graticules that are approximately parallel to the ground (i.e., graticules with very small angles), and retain areas that are obviously tilted or vertical, i.e., areas where retaining walls may exist, to obtain non-ground point clouds. Extracted non-ground point clouds such as Figure 6As shown.

[0038] S5. Cluster the extracted non-ground point cloud (e.g., DBSCAN clustering). The clustering result is as follows: Figure 7 As shown, principal component analysis (PCA) is used to calculate the eigenvalues ​​of each cluster to obtain the linearity of each cluster. Several clusters with high linearity are selected as candidate clusters, and each candidate cluster is scored. The cluster with the highest score is considered the retaining wall point cloud. In this embodiment of the invention, the three point cloud clusters with the highest linearity are selected as candidate clusters. The selected retaining wall point cloud is shown below. Figure 8 As shown.

[0039] Each candidate cluster is scored, including: boundary identification of the expanded working area point cloud, defining a safe boundary safe_bound / GridSize, i.e. reserving the width of the safe_bound boundary area, traversing all grids in each candidate cluster, and determining whether the grid coordinates of the grid are within the boundary area.

[0040] For each candidate cluster, the percentage of grid cells in the boundary region relative to the total number of grid cells in the candidate cluster is used as the boundary coverage rate. A higher boundary coverage rate indicates that the candidate cluster is closer to the scene edge. Considering that retaining walls are generally located at the boundary of the working area, boundary coverage rate is used as an important indicator for scoring the clustered point cloud. The specific scoring method is as follows: ; Where `edge_coverage` is the edge coverage of the candidate cluster, representing the proportion of points in the point cloud of the candidate cluster that belong to the working region boundary within the entire candidate cluster. `liners` represents the principal scale length feature of the candidate cluster, i.e., the feature value in the principal direction. `sum_line` is the sum of the lengths of all candidate clusters. `linearity` represents the linearity of the candidate cluster; the larger this value, the more the cluster tends to be a long, thin straight line. `class_z` is the average height of the candidate cluster, and `sum_z_average` is the average of the average heights of all candidate clusters, representing the global average height.

[0041] This formula scores each candidate cluster and selects a cluster that is "tall, long, slender, and close to the edge," which matches the characteristics of a retaining wall.

[0042] S6, perform height segmentation again on the extracted retaining wall point cloud, such as... Figure 9As shown, the height segment with the most point cloud points is found and a straight line fitting operation is performed to obtain the straight line equation of the retaining wall. Based on the straight line equation of the retaining wall, the straight line of the retaining wall is obtained through linear interpolation, and the straight line of the retaining wall is translated upward to fit the top line, thereby realizing the recognition of the retaining wall. In this embodiment of the invention, the straight line fitting method is the RANSAC random sampling method.

[0043] Specifically, since radar can only scan the surface of the retaining wall, and the top of the retaining wall is uneven with poor straight-line characteristics, while the bottom has too much contact with the ground point cloud, the fitting process is easily affected. Combining the installation height of the lidar, it is determined that the middle part of the retaining wall point cloud has the largest number of points and the best linear characteristics. Therefore, performing a straight-line fitting on this part yields more accurate results. In this embodiment of the invention, the height segment with the largest number of points is taken as the intermediate layer. A straight-line fitting is performed on the point cloud of the intermediate layer to obtain the straight-line equation of the retaining wall. The relative position of the centroid of the point cloud above the intermediate layer and the centroid of the intermediate layer is calculated. The straight line of the retaining wall is then shifted upwards by this relative position to conform to the top line.

[0044] During interpolation, the span of the point cloud on the straight line equation of the retaining wall along the y-axis and x-axis is calculated separately. If the point cloud is longer along the y-axis than along the x-axis, the y-axis is chosen as the principal axis. This is because if a straight line is long in the y-direction and short in the x-direction, a large interpolation step along the x-axis would result in sparse points. Therefore, the longer axis is chosen as the principal axis to ensure a uniform distribution of interpolation points. The principal axis direction is determined by whether it is the x-axis or the y-axis. Based on the fitted straight line equation of the retaining wall, interpolation is performed along the principal axis to obtain the linear characteristic line of the retaining wall. The final fitted point cloud of the retaining wall straight line is shown below. Figure 10 As shown.

[0045] Example 2 This invention provides an automatic identification system for retaining walls at construction sites based on point cloud maps, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the automatic identification method for retaining walls at construction sites based on point cloud maps in Embodiment 1 above.

[0046] The electronic device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor performs various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.

[0047] The relevant technical solutions are the same as above, and will not be repeated here.

[0048] Example 3 This invention provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the automatic identification method for retaining walls at construction sites based on point cloud maps in Embodiment 1 above.

[0049] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0050] The relevant technical solutions are the same as above, and will not be repeated here.

[0051] Example 4 This invention provides a computer program product, including a computer program that, when run on a computer, causes the computer to perform the steps of the automatic identification method for retaining walls at construction sites based on point cloud maps in Embodiment 1 above.

[0052] The relevant technical solutions are the same as above, and will not be repeated here.

[0053] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic identification of retaining walls at construction sites based on point cloud maps, characterized in that, include: Obtain the point cloud coordinates of the bulldozer working area in the point cloud map of the construction site, and then convert the point cloud coordinates to a standard coordinate system and perform spatial rasterization to obtain a raster map. The raster map is divided into multiple height segments according to the point cloud height. The height segment with the most grid cells is selected as the ground segment. Grid cells in the height segment whose point cloud height is greater than the point cloud height of the ground segment are selected as candidate grid cells. The angle between the average normal direction of the candidate grid cells and the ground is determined. Candidate grid cells with an angle less than a preset threshold are deleted. The point clouds in the remaining candidate grid cells are selected as non-ground point clouds. For the non-terrestrial point cloud clustering, principal component analysis is used to calculate the linearity of each cluster, and several clusters with high linearity are selected as candidate clusters. Each candidate cluster is scored based on its boundary coverage rate, which is proportional to the score of the candidate cluster. The candidate cluster with the highest score is selected as the retaining wall point cloud. The boundary coverage rate is the ratio of the number of grid cells in the boundary region of the bulldozer working area in the candidate cluster to the total number of grid cells in the candidate cluster, and this ratio is proportional to the score of the candidate cluster. Retaining wall identification is performed based on the point cloud of the retaining wall.

2. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to claim 1, characterized in that, Retaining wall identification based on the retaining wall point cloud includes: The retaining wall point cloud is further segmented by height. The height segment with the most point cloud points is fitted with a straight line to obtain the straight line equation of the retaining wall. The straight line equation of the retaining wall is linearly interpolated to obtain the straight line of the retaining wall. The straight line of the retaining wall is then translated upward to fit the top line, thereby realizing the recognition of the retaining wall.

3. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to claim 2, characterized in that, The retaining wall is moved upwards in a straight line to fit the top line, including: Calculate the relative position of the centroid of the point cloud above the height segment with the most point cloud content to the centroid of the point cloud above the height segment with the most point cloud content, and then translate the retaining wall upwards in a straight line to fit the top line. Linear interpolation is performed on the straight line equation of the retaining wall, including: Calculate the difference between the maximum and minimum values ​​of the point cloud on each coordinate axis of the retaining wall straight line equation; select the coordinate axis with the larger difference as the main axis, and perform point interpolation along the main axis to obtain the retaining wall straight line.

4. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to any one of claims 1-3, characterized in that, Calculating the average normal direction of the candidate grid includes: Traverse all points within each candidate grid, search for the k nearest neighbors of each point, perform principal component analysis on the k nearest neighbors to obtain the normal of the point; unify the orientation of the normal directions of all points within the candidate grid, and take the weighted average of the normals of each point and its k nearest neighbors as the normal vector of each point; add the normal vectors of all points within the candidate grid and normalize them to obtain the average normal direction of the candidate grid.

5. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to claim 1, characterized in that, The candidate clustering score is calculated as follows: in, The candidate cluster score is defined by edge_coverage, which represents the edge coverage of the candidate cluster, liners represents the principal scale length feature of the candidate cluster, sum_line represents the sum of the lengths of all candidate clusters, linearity represents the linearity of the candidate cluster, class_z represents the average height of the candidate cluster, and sum_z_average represents the average height of all candidate clusters.

6. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to claim 1, characterized in that, Obtaining the raster map includes: Calculate the number of grid cells in the x and y axes in the standard coordinate system; Calculate the mean and variance of the point cloud height within each grid cell, and use the mean as the point cloud height of the corresponding grid cell and the variance as the variance of the corresponding grid cell. The average variance of the point cloud heights of all grid cells is calculated based on the point cloud height of each grid cell. If the difference between the variance of a grid cell and the average variance is greater than a preset threshold, the grid cell is removed as an outlier, and the grid map with outliers removed is obtained.

7. The method for automatic identification of retaining walls at construction sites based on point cloud maps according to claim 1, characterized in that, The bulldozer's working area is the expanded working area; Obtaining the point cloud coordinates of the expanded working area includes: The expansion factor is calculated from the point cloud map based on the vertex coordinates of each vertex in the provided original working area. : ;in, The distance from each vertex of the original working region to the centroid of the original working region is given by expansion_distance, where expansion_distance is the set expansion distance. The expansion factor Multiply by the x and y coordinates of each vertex to obtain the coordinates of each vertex in the expanded working area; The point cloud map is subjected to intensity filtering and outlier filtering. Based on the coordinates of each vertex in the expanded working area and the filtered point cloud map, the point cloud coordinates within the expanded working area are obtained.

8. An automatic identification system for retaining walls at construction sites based on point cloud maps, characterized in that, Includes computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the automatic identification method for retaining walls at construction sites based on point cloud maps as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the automatic identification method for retaining walls at construction sites based on point cloud maps as described in any one of claims 1-7.

10. A computer program product, characterized in that, The method includes a computer program that, when run on a computer, causes the computer to perform the automatic identification method for retaining walls at construction sites based on point cloud maps as described in any one of claims 1-7.

Images