Method for estimating particle size of block stone based on laser radar

Abstract

The invention discloses a block stone particle size estimation method based on a laser radar, and the method comprises the following steps: obtaining the original point cloud data of a target riprap ship, and extracting the point cloud data of a block stone region from the original point cloud data based on cloth simulation filtering; obtaining a block stone depth map according to the block stone area point cloud data; performing image enhancement processing on the block stone depth map to obtain a block stone enhancement image; determining a block stone edge area of the block stone enhancement image based on the inter-class distance; segmenting the block stone in the block stone enhancement image based on the block stone edge region to obtain a block stone point cloud segmentation result; and calculating the particle size of the block stone according to the block stone point cloud segmentation result. According to the method, accurate segmentation and particle size estimation of the block stone can be realized, and the influence of overlapping and irregularity in a complex rock heap is reduced.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and 3D laser scanning technology, specifically relating to a method for estimating the particle size of stones based on lidar. Background Technology

[0002] Underwater riprap placement is a crucial component of waterway improvement projects, serving as a primary means of bank protection, foundation stabilization, and flow guidance. Statistics show that riprap placement accounts for 35%-45% of the total investment in typical waterway improvement projects in the middle and lower reaches of the Yangtze River, and its construction quality directly impacts the effectiveness of waterway improvement. The size of the riprap is a core element in quality control during riprap placement, a key factor in balancing project quality, safety, and economy, and a crucial basis for subsequent quality traceability. Traditionally, riprap size estimation relied on manual methods. During riprap placement, workers would assess the approximate size and uniformity of the riprap from incoming vessels based on experience, or randomly sample a certain number of stones. The results were then analyzed from all sample stones. This traditional method of underwater riprap size estimation has significant limitations and uncertainties; it is not only time-consuming and labor-intensive, but also lacks accuracy and objectivity.

[0003] With the development of automation and data processing technologies, in order to improve the accuracy and efficiency of boulders particle size values, patent CN115046473A discloses a quality control method for underwater boulders based on three-dimensional point clouds. It uses the AABB bounding box method to obtain the particle size value of the boulders. However, the bounding box size (length, width, height) calculated by the AABB bounding box method is usually larger than the actual effective size of the boulders in that direction, which means that the obtained boulders particle size value will contain a large amount of "invalid space" that does not belong to the boulders themselves, resulting in a large error in the measurement of boulders particle size. Moreover, the patent uses a handheld three-dimensional laser scanner, which still requires professional personnel to board the ship to collect data. The post-processing steps are complicated and time-consuming, and it is impossible to realize real-time monitoring and automated analysis at the construction site.

[0004] While UAV-based LiDAR (Light Detection and Ranging) measurements are efficient and have automation potential, they also face challenges in large-scale UAV-acquired point cloud data. Issues such as large variations in block size, spatial overlap, and uneven point density within the point cloud data hinder reliable block segmentation and limit accurate granularity estimation. Therefore, improving the accuracy of block granularity estimation is crucial for further expanding the engineering applications of UAV LiDAR scanning. Summary of the Invention

[0005] To address the problem of inaccurate estimation of stone size in trebuchets using existing technologies, this invention provides the following technical solution:

[0006] A method for estimating the particle size of boulders based on lidar is provided, which includes the following steps:

[0007] The original point cloud data of the target trebuchet is acquired based on lidar, and the point cloud data of the boulder area is extracted from the original point cloud data based on cloth simulation filtering.

[0008] Obtain a depth map of the boulders based on the point cloud data of the boulder area;

[0009] The depth map of the boulders is subjected to image enhancement processing to obtain an enhanced image of the boulders;

[0010] The edge region of the block in the block enhancement map is determined based on the inter-class distance;

[0011] The blocks in the enhanced block map are segmented based on the edge region of the blocks to obtain the block point cloud segmentation result Ps;

[0012] Furthermore, the particle size of the stone is calculated based on the point cloud segmentation result Ps of the stone.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] This invention utilizes the characteristics of rock pile accumulation and the distinct depth edges formed by height variations between rocks to assist in rock segmentation, achieving precise rock segmentation and effectively mitigating the impact of overlap and irregularity in complex rock piles. Specifically, during automatic estimation, a customized control filter is used to obtain a rock pile enhancement map, and region growing is performed using the edge information provided by the enhancement map as a constraint. This achieves precise rock pile segmentation and grain size estimation, reducing the impact of overlap and irregularity in complex rock piles. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the method for estimating the particle size of boulders based on lidar in this invention.

[0016] Figure 2 Point cloud of the rocky area extracted in this invention;

[0017] Figure 3 This is a diagram showing the depth of the boulders in an embodiment of the present invention;

[0018] Figure 4 This is a diagram showing the enhancement of the stone block in an embodiment of the present invention;

[0019] Figure 5 This is a diagram of the edge region of the stone block in an embodiment of the present invention;

[0020] Figure 6 Ps represents the point cloud segmentation result of the block stone in this embodiment of the invention;

[0021] Figure 7 This is a histogram showing the distribution of stone particle size in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1:

[0024] like Figure 1 As shown, this embodiment provides a method for estimating the particle size of boulders based on lidar, including the following steps:

[0025] S1. Obtain the original point cloud data of the target trebuchet, and extract the point cloud data of the boulder region P{p1(x1,y1,z1,rgb1),p2(x2,y2,z2,rgb2),…,p n (x n ,y n ,z n ,rgb n )}(like Figure 2 As shown), where p n (x n ,y n ,z n ,rgb n ) represents the point cloud data of the nth stone region; x, y, z represent the three-dimensional coordinate information of the point cloud of the stone region, rgb represents the color value of the point cloud, and n is the number of original point cloud data; in this embodiment, the original point cloud data is acquired by the lidar carried by the UAV, and the stone-throwing ship is a ship loaded with stones;

[0026] S2. Obtain the depth map of the boulders based on the point cloud data of the boulder area, which specifically includes the following steps:

[0027] The point cloud data of the boulders region is orthographically projected in the xy direction at a preset grid resolution (e.g., 0.1m) to convert the height z of each boulder region's point cloud data into its grayscale value gray, thereby obtaining the boulder depth map Pimage{pixel(col,row,gray)} (e.g., ...). Figure 3 As shown in the figure, col and row are the column coordinates and row coordinates of any pixel in the Pimage depth map, respectively.

[0028] S3. Perform image enhancement processing on the depth map of the boulders to obtain, as shown in the image... Figure 4 The enhanced image of the boulders shown in this embodiment demonstrates how image enhancement processing strengthens the edges of boulders of different sizes in the depth map, making their edges more distinct while reducing the impact of noise.

[0029] S4. Determine the edge region of the block enhancement map based on the inter-class distance;

[0030] S5. Segment the blocks in the enhanced block map based on the edge region of the blocks to obtain the block point cloud segmentation result Ps;

[0031] And, S6, calculate the particle size of the stone based on the point cloud segmentation result Ps of the stone.

[0032] Specifically, in step S1, extracting the point cloud data of the rocky area from the original point cloud data based on cloth simulation filtering includes the following steps:

[0033] S11. Flip the original point cloud data so that the bottom surface faces up to obtain flipped point cloud data, so as to simulate the falling of the fabric.

[0034] S12. Construct a cloth mesh, wherein the cloth mesh nodes are (x ij ,y ij ,z ij The grid resolution is 0.1m, and the initial position of the cloth grid is horizontally placed above the highest point in the flipped point cloud data to complete the initialization of the cloth grid;

[0035] S13. Determine the height z of the point cloud with the smallest z-coordinate among the flipped point cloud data located within the neighborhood of each node of the fabric mesh. target ;

[0036] S14. Update the height of the current node in the cloth mesh based on the following formula until the height of the current node is less than or equal to the height z. target Stop updating height at this time;

[0037]

[0038] in, This represents the height of the current node at coordinates (i,j) after the t-th iteration. The height of the current node at coordinates (i,j) after the (t+1)th iteration is represented; g represents the simulated gravity step size, which ranges from 0.1 to 1.0 m / iteration step. The convergence threshold has a value range of 1×10. -4 ~1×10-2 ;

[0039] Therefore, in this embodiment, the height is calculated iteratively using the ε value, so that the height of the fabric nodes is updated gradually until the maximum height change of all nodes is less than the convergence threshold ε, at which point the fabric falling process terminates.

[0040] S15. Repeat steps S13-S14 above, so that the updated height of each node in the cloth mesh is less than or equal to the height z. target ;

[0041] S16. Calculate the height difference between each node and the point cloud in the flipped point cloud data after the height update according to the following formula. And the bottom point is determined based on the height difference Δz:

[0042]

[0043] in, Indicates the height difference, z point Let z represent the z-coordinate of any point cloud in the flipped point cloud data. cloth This is represented as the z-coordinate of the cloth mesh at the same (x, y) position; further, if Δz < <T z If the node is considered to be a bottom point, then in this embodiment, Δz < <T z , and |zz m |<T z The node with a value of 2 is taken as the bottom point, T z The height difference threshold is defined as T, with a value ranging from 0.05 to 0.2 m. In this embodiment, T is used. z =0.1m;

[0044] S17. Determine convex hull point cloud data in the flipped point cloud data using the convex hull algorithm. The convex hull point cloud data includes all flipped point cloud data within the convex hull with the bottom point as the boundary.

[0045] S18. Clustering calculations are performed on the convex hull point cloud data based on the Euclidean Cluster Extraction algorithm, and the connected cluster with the most point clouds is taken as the point cloud data of the boulders region retained after removing the point cloud from the bottom of the trebuchet. This specifically includes the following steps:

[0046] Calculate the Euclidean distance between point clouds in the convex hull point cloud data, and based on the Euclidean distance, merge point clouds that satisfy the neighborhood distance constraint to obtain several connected clusters, denoted as {C1, C2, ..., C...}. n}; where C nThe nth connected cluster is represented, where each of the connected subsets corresponds to a spatially independent region or physical object in the flipped point cloud data.

[0047] The connected cluster C with the largest number of point clouds will be selected. max As point cloud data for the rocky area.

[0048] Therefore, this embodiment uses fabric simulation to generate a hanging fabric surface under the action of gravity, and simulates the process of the fabric falling to the ground on the surface of the original point cloud data to obtain the bottom point cloud data; then, the bottom point cloud data is removed from the original point cloud data by using the convex hull algorithm and Euclidean clustering to obtain the point cloud of the block area, which provides more accurate point cloud data for the estimation of block particle size.

[0049] Furthermore, in step S3, the image enhancement processing of the rock depth map includes the following steps:

[0050] Construct the Hessian matrix, its expression is as follows:

[0051]

[0052] In the formula, Gxx(x), Gxy(x), Gyx(x), and Gyy(x) are all second-order partial derivatives of the Gaussian function; I(x) is the rock depth map.

[0053] Let λ1 and λ2 be the first and second eigenvalues ​​of the Hessian matrix, respectively, and |λ2| ≥ |λ1|. Based on the first eigenvalue λ1 and the second eigenvalue λ2, construct the eigenline structure description term R. B Intensity description item S:

[0054]

[0055]

[0056] Wherein, the linear structure description term R B R is used to describe the linear structural features of a local region. B The larger the value, the closer the local area is to a linear structure; the intensity description term S is used to enhance the high gray value features of the layer boundary, making the features more prominent;

[0057] A filter is constructed, and the depth map of the boulders is filtered based on the following formula to obtain the following result: Figure 4 The formula for the stone reinforcement diagram shown is as follows:

[0058]

[0059] Where IM(δ0) represents the scale parameter of the Gaussian function. At that time, the enhanced image of the stone block obtained after filtering; κ and γ are the filter's description of the linear structure R. B The sensitivity threshold for the intensity description term S.

[0060] Since stones of different sizes have obvious edges in the depth map, in this embodiment, the local features of the depth map can be determined based on the magnitude of the first feature value λ1 and the second feature value λ2. For example, if both λ1 and λ2 are small, the local area is a flat gray area; if λ1 is small and λ2 is large, the local area has a small gray change in one direction and a significant change in another orthogonal direction, indicating that the local area has a linear edge structure; if both λ1 and λ2 are large, it indicates that the local structure is a corner structure.

[0061] Furthermore, since there is a significant difference between the middle position and the two sides of the second-order partial derivative of the Gaussian kernel, and the gray value near the edge of the stone depth map also changes according to the trend of weak to strong and then weak again, the second-order partial derivative of the Gaussian kernel and the edge structure of the stone depth map are similar. Moreover, since the edge structure and the signal of the second-order partial derivative of the Gaussian kernel are also similar, in this embodiment, when performing the convolution operation of multi-scale linear structure enhancement filtering, the position of the edge structure in the stone depth map will obtain a higher response value, thereby being enhanced, and thus realizing the enhancement of the stone depth map.

[0062] Furthermore, step S4, determining the edge region of the block enhancement map based on inter-class distance, includes the following steps:

[0063] S41. Set a grayscale value threshold G, and classify the pixels in the enhanced block image whose grayscale value is less than the grayscale value threshold G into the first category, and classify the pixels in the enhanced block image whose grayscale value is greater than or equal to the grayscale value threshold G into the second category.

[0064] S42. Binarize the enhanced image of the boulders based on the following formula to obtain the class spacing D:

[0065]

[0066] D is the inter-class distance; μ1 and μ2 are the mean gray values ​​of the first class and the second class, respectively; G is the gray value threshold, G∈[0,255];

[0067] S43. Traverse the value range [0,255] to take different values ​​for the gray value threshold G, and repeat step S42 based on the different gray value threshold G values ​​to obtain the inter-class spacing D corresponding to the different gray value threshold G values.

[0068] S44. The grayscale threshold G corresponding to the maximum inter-class spacing D is used as the distinction threshold between edge regions and non-edge regions in the enhanced block image. Pixels in the enhanced block image with grayscale values ​​greater than or equal to the distinction threshold are defined as edge points, and the grayscale value of each edge point is assigned to 255, thereby determining... Figure 5 The area at the edge of the boulders shown.

[0069] Therefore, in this embodiment, the enhanced image of the stone block can be binarized using the maximum inter-class distance method, making the edge region clearer.

[0070] Furthermore, in step S5, the blocks in the enhanced block image are segmented based on the block edge region to obtain the block point cloud segmentation result Ps, including the following steps:

[0071] S51. Determine the outline of each stone based on the edge region of the stone block, and the outline of each stone is a closed structure to prevent excessive merging of adjacent clusters.

[0072] S52. Perform a neighborhood search within the area enclosed by the current block outline according to the preset radius r0 to determine the neighborhood points of each pixel within the area enclosed by the current block outline.

[0073] S53. Create an empty seed set, and select one from the neighboring points as the initial seed point P0, and add it to the seed set to obtain a single-point seed set;

[0074] S54. Using the initial seed point P0 as the center, perform a neighborhood search within the enhanced image of the block with a radius r1, and select the neighboring points within the neighborhood as candidate points P corresponding to the current initial seed point P0. i And calculate the initial seed point P0 and each candidate point P i The positional difference d; preferably, the radius r0 is 0.5~0.8m;

[0075] If d≤T d And the candidate point P i If it is not an edge point, then the candidate point P is... i Added to the single-point seed set, thus completing the candidate point P. i The classification; the position difference threshold T d It is 0.5m;

[0076] Repeat the above steps until the classification of all candidate points of the current initial seed point P0 is completed. Then, take the single-point seed set as the clustering result of the current initial seed point P0, and mark the current initial seed point P0 and all its corresponding candidate points as classified points.

[0077] S55. Repeat steps S52 to S54 to obtain the clustering results of all initial seed points within the closed area of ​​the current block outline, and use them as the clustering results of the current block.

[0078] S56. Repeat step S55 to obtain the clustering results for each stone, and assign a corresponding color to each clustering result to obtain a stone category segmentation map.

[0079] S57. Assign each pixel in the block category segmentation map to a point cloud in the block region point cloud, and assign each point cloud the same color as its corresponding pixel to obtain m point cloud clusters C1, C2, ..., Cm;

[0080] In this embodiment, the correspondence between the pixel and the point cloud of the block area can be determined based on the coordinate transformation relationship between the pixel and the point cloud;

[0081] S58. Calculate the horizontal distance between the unclassified points in the point cloud of the rocky area and each point cloud cluster.

[0082] If the height of the unclassified point is lower than the average height of the current point cloud cluster, and the distance between the unclassified point and the current point cloud cluster is the smallest, then the unclassified point is classified into the current point cloud cluster.

[0083] S59. Repeat step S58 until all unclassified points are classified, thereby obtaining the final block point cloud segmentation result Ps. At this point, each item in the block point cloud segmentation result Ps corresponds to a block point cloud cluster. Figure 6 Each colored block represents a cluster of rocks and clouds.

[0084] Therefore, this embodiment establishes a closed contour structure around a single block based on the edges extracted from the block edge clustering map to prevent excessive merging of adjacent clusters. At the same time, a region growing algorithm is used and edge constraints are considered to accurately divide the block region point cloud obtained in step S2 into multiple block point cloud clusters.

[0085] Furthermore, in step S6, the particle size of the stone is calculated based on the stone point cloud segmentation result Ps, including the following steps:

[0086] Principal component analysis was performed on each cluster of point clouds in the segmented point cloud of the block stone, Ps, to calculate the lengths of several (e.g., 3) principal axes of each cluster. The arithmetic mean of all principal axis lengths was used as the equivalent grain size of the block stone, which was then used as the block stone size. A grain size distribution histogram was constructed based on the estimated grain size of the block stone. The results are as follows: Figure 7 As shown.

[0087] In summary, this invention removes the point cloud from the bottom of the quarry by fabric simulation filtering, providing more accurate point cloud data for estimating the particle size of the boulders. Furthermore, based on the characteristics of the boulder accumulation, a dedicated control filter is designed to obtain an enhanced boulder map within the accurate boulder accumulation area, which enhances the edges of the boulders. This enhanced boulder map allows for precise segmentation of the boulders in the quarry area, effectively mitigating the impact of overlap and irregularities in complex rock piles. Simultaneously, region growing is performed using the edge information provided by the enhanced boulder map as a constraint, enabling precise segmentation and particle size estimation of the boulders.

[0088] It should be noted that, in this document, terms such as "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for estimating the particle size of boulders based on lidar, characterized in that, Includes the following steps: The original point cloud data of the target trebuchet is acquired based on lidar, and the point cloud data of the boulder area is extracted from the original point cloud data based on cloth simulation filtering. Obtain a depth map of the boulders based on the point cloud data of the boulder area; The depth map of the boulders is subjected to image enhancement processing to obtain an enhanced image of the boulders; The edge region of the block in the block enhancement map is determined based on the inter-class distance; The blocks in the enhanced block map are segmented based on the edge region of the blocks to obtain the block point cloud segmentation result Ps; Furthermore, the particle size of the stone is calculated based on the point cloud segmentation result Ps of the stone.

2. The method for estimating the particle size of boulders as described in claim 1, characterized in that, Extracting point cloud data of the rocky area from the original point cloud data based on cloth simulation filtering includes the following steps: The original point cloud data is flipped so that the bottom surface is facing up to obtain flipped point cloud data; Construct a cloth mesh and place the initial position of the cloth mesh horizontally above the highest point in the flipped point cloud data to complete the initialization of the cloth mesh; Determine the height z of the point cloud with the smallest z-coordinate among the flipped point cloud data located within the neighborhood of each node of the fabric mesh. target ; The height of the current node in the cloth mesh is updated based on the following formula until the height of the current node is less than or equal to the height z. target Stop updating height at this time; ； in, This represents the height of the current node at coordinates (i,j) after the t-th iteration. The height of the current node at coordinates (i,j) is updated after the (t+1)th iteration; g represents the simulated gravity step size. This is the convergence threshold; Repeat the above steps until the updated height of each node in the cloth mesh is less than or equal to the height z. target ; The height difference Δz between each node and the point cloud in the flipped point cloud data is calculated according to the following formula after the height update, and the bottom point is determined based on the height difference Δz: ；； Among them, z point Let z represent the z-coordinate of any point cloud in the flipped point cloud data. cloth Let Δz represent the z-coordinate of the cloth mesh at the same (x, y) position; and if Δz < <T z If the node is not a bottom point, then it is considered to be a bottom point. The convex hull point cloud data is determined in the flipped point cloud data using the convex hull algorithm. The convex hull point cloud data includes all flipped point cloud data within the convex hull with the bottom point as the boundary. The convex hull point cloud data is clustered based on the Euclidean clustering algorithm, and the connected cluster with the most point clouds is taken as the point cloud data of the boulders region after removing the point cloud at the bottom of the trebuchet.

3. The method for estimating the particle size of boulders as described in claim 2, characterized in that, This will satisfy Δz< <T z , and |zz m |<T z The node with a value of 2 is taken as the bottom point, T z The height difference threshold ranges from 0.05 to 0.2m.

4. The method for estimating the particle size of boulders as described in claim 1, characterized in that, The process of obtaining a depth map of the boulders based on the point cloud data of the boulder area includes the following steps: The point cloud data of the rock area is orthographically projected in the xy direction with a preset grid resolution to convert the height z of the point cloud data of each rock area into the grayscale value of the point cloud data, so as to obtain the rock depth map.

5. The method for estimating the particle size of boulders as described in claim 1, characterized in that, Image enhancement processing of the rock depth map includes the following steps: Construct the Hessian matrix, its expression is as follows: ； In the formula, Gxx(x), Gxy(x), Gyx(x), and Gyy(x) are all second-order partial derivatives of the Gaussian function; I(x) is the rock depth map. Let λ1 and λ2 be the first and second eigenvalues ​​of the Hessian matrix, respectively, and |λ2| ≥ |λ1|. Based on the first eigenvalue λ1 and the second eigenvalue λ2, construct the eigenline structure description term R. B Intensity description item S: ； ； A filter is constructed, and the depth map of the boulders is filtered based on the following formula to obtain an enhanced boulders map: ； Where IM(δ0) represents the scale parameter of the Gaussian function. At that time, the enhanced image of the stone block obtained after filtering; κ and γ are the filter's description of the linear structure R. B The sensitivity threshold for the intensity description term S.

6. The method for estimating the particle size of boulders as described in claim 1, characterized in that, Determining the edge region of the block enhancement map based on inter-class distance includes the following steps: Set a grayscale value threshold G, and classify the pixels in the enhanced block image whose grayscale value is less than the grayscale value threshold G into the first category, and classify the pixels in the enhanced block image whose grayscale value is greater than or equal to the grayscale value threshold G into the second category; The enhanced image of the boulders is binarized using the following formula to obtain the class spacing D: ； D is the inter-class distance; μ1 and μ2 are the mean gray values ​​of the first and second classes, respectively; G is the gray value threshold. Iterate through the range of grayscale threshold G to obtain different values ​​for grayscale threshold G, and repeat the above steps based on the different values ​​of grayscale threshold G to obtain the inter-class spacing D corresponding to the different values ​​of grayscale threshold G. The gray value threshold G corresponding to the maximum inter-class spacing D is used as the threshold for distinguishing edge regions and non-edge regions in the enhanced block image. In the enhanced block image, pixels with gray values ​​greater than or equal to the threshold are determined as edge points, and the gray value of each edge point is assigned to 255.

7. The method for estimating the particle size of boulders as described in claim 6, characterized in that, The grayscale threshold G ranges from [0, 255].

8. The method for estimating the particle size of boulders as described in claim 1, characterized in that, The blocks in the enhanced block map are segmented based on the edge regions of the blocks to obtain the block point cloud segmentation result Ps, including the following steps: The outline of each stone is determined based on the edge region of the stone block, and the outline of each stone is a closed structure. A neighborhood search is performed within the area enclosed by the current block outline according to a preset radius r0 to determine the neighborhood points of each pixel within the area enclosed by the current block outline. Create an empty seed set, and select one from the neighborhood points as the initial seed point P0, and add it to the seed set to obtain a single-point seed set; Using the initial seed point P0 as the center, a neighborhood search is performed within the enhanced image of the block, and the neighboring points within the neighborhood are used as candidate points P corresponding to the current initial seed point P0. i And calculate the initial seed point P0 and each candidate point P i The positional difference d; if d≤T d And the candidate point P i If it is not an edge point, then the candidate point P is... i Add to the single-point seed set to complete the candidate point P. i Classification; Repeat the above steps until the classification of all candidate points of the current initial seed point P0 is completed. Then, take the single-point seed set as the clustering result of the current initial seed point P0, and mark the current initial seed point P0 and all its corresponding candidate points as classified points. Repeat the above steps to obtain the clustering results of all initial seed points within the closed area of ​​the current block outline, and use them as the clustering results of the current block. Repeat the ascending steps to obtain the clustering results for each stone, and assign a corresponding color to each clustering result to obtain a stone category segmentation map; Each pixel in the block category segmentation map is mapped to a point cloud in the block region point cloud, and each point cloud is assigned the same color as its corresponding pixel to obtain several point cloud clusters. Calculate the horizontal distance between the unclassified points in the point cloud of the blocky area and each point cloud cluster. If the height of the unclassified point is lower than the average height of the current point cloud cluster, and the distance between the unclassified point and the current point cloud cluster is the smallest, then the unclassified point is assigned to the current point cloud cluster. Repeat the above steps until all unclassified points are classified to obtain the block point cloud segmentation result Ps.

9. The method for estimating the particle size of boulders as described in claim 8, characterized in that, The radius r0 is 0.5~0.8m.

10. The method for estimating the particle size of boulders as described in claim 1, characterized in that, The calculation of the particle size of the block stone based on the block stone point cloud segmentation result Ps includes the following steps: performing principal component analysis on each block stone point cloud cluster in the block stone point cloud segmentation result Ps to calculate several principal axis lengths of each block stone point cloud cluster, and using the arithmetic mean of all principal axis lengths as the equivalent particle size of the block stone, which is used as the block stone size.

Images