CN-122023281-A - Three-dimensional point cloud-based rock-fill material grading detection method

CN122023281ACN 122023281 ACN122023281 ACN 122023281ACN-122023281-A

Abstract

The invention discloses a three-dimensional point cloud-based rock-fill material grading detection method, and relates to the technical field of rock-fill material grading detection. The method comprises the steps of obtaining three-dimensional point cloud data of the stone particles after paving through laser scanning, projecting the three-dimensional point cloud data along a overlook direction to generate a depth map, inputting the depth map into an example segmentation model, extracting segmentation masks of the particles, realizing accurate segmentation of the three-dimensional particles by mapping the masks back to point cloud, further extracting three-dimensional morphological characteristics of each particle, predicting the volume and the particle size of the particles based on the characteristics, and finally combining the particle size and volume information of the particles to complete grading statistics and grading curve drawing of the stone pile. The invention can efficiently and accurately realize the grading detection of the stacking material and has the advantages of high precision, strong automation degree and the like.

Inventors

WU YILONG
YUAN KUN
DU TAO
HUANG GUOLIANG
YAO QIANG
LI HONGTAO
LI CHUNQUAN
Liao Yabin
LIU JINGBIAO
JI PENG
SHI MENGNAN
ZENG JUN

Assignees

中国水利水电第七工程局有限公司
四川大学

Dates

Publication Date: 20260512
Application Date: 20260108

Claims (6)

1. The rock-fill material grading detection method based on the three-dimensional point cloud is characterized by comprising the following steps of: S1, collecting stone point cloud data; S2, projecting the rock-fill material point cloud along the overlooking direction to generate a corresponding depth image; S3, inputting the depth map into an example segmentation model to obtain a segmentation mask of each stone particle; S4, mapping the segmentation mask back to a three-dimensional point cloud, and extracting three-dimensional morphological characteristics of each particle; s5, predicting the volume and the particle size based on the three-dimensional characteristics of the particles, and drawing a grading curve.
2. The three-dimensional point cloud-based rock-fill grading detection method according to claim 1, wherein the step S2 comprises: S21, rotating the stacking material point cloud to a standard horizontal coordinate system so as to be subjected to depth projection; Firstly, carrying out plane fitting on point clouds by adopting a random sampling consistency RANSAC algorithm, extracting ground normal vectors from the point clouds, wherein for any point P i , the distance from the point P i to a fitting plane is as follows: ; When Di < epsilon (preset threshold), P i is regarded as the inner point of the current plane, and the RANSAC algorithm finally determines the ground normal vector by iteratively searching the plane with the largest inner point number Then, the point cloud is rotated to a horizontal coordinate system by using a Rodrigues rotation formula, and a rotation matrix R 2 is: ; ; ; S22, projecting the rotated point cloud to an XOY plane along a Z axis, and constructing a regular two-dimensional grid; setting the size of the grid cell to be delta x delta y, traversing all points, and determining the index position of the points in the grid according to X, Y coordinates of the points: ; In each grid unit, adopting a maximum height projection principle, and reserving the point with the maximum Z value in the area as a depth value so as to generate a corresponding depth matrix: ; The depth matrix D is converted into a gray level diagram form after normalization processing, a corresponding two-dimensional depth diagram is generated, and the gray level value of each pixel reflects the surface height information of the area and is used as the input of a subsequent example segmentation model.
3. The three-dimensional point cloud-based rock-fill grading detection method according to claim 1, wherein the example segmentation model preferably adopted in the step S3 is YOLOv-seg.
4. The three-dimensional point cloud based rockfill material grading detection method according to claim 1, wherein the step S4 comprises: s41, after obtaining the depth map segmentation mask, back projecting the depth map mask back to the point cloud mask, wherein for any pixel position in the depth map And depth value The corresponding point cloud coordinate calculation formula is as follows: ; s42, extracting three-dimensional features of each divided particle point cloud, and calculating three-dimensional geometric features such as a long axis, a secondary long axis, a short axis, a surface area, a volume, sphericity, roundness and the like; 1) Calculating the main scale direction of the particles by adopting a main component analysis method; The main scale of the rock particle is generally formed by three mutually perpendicular axial dimensions, namely a long axis L, a secondary long axis W and a short axis T, the principal component analysis method is adopted to calculate a point cloud covariance matrix and solve a characteristic value and a characteristic vector of the point cloud covariance matrix, and the point cloud data of the particle is projected to the direction with the maximum variance, so that a projection coordinate system is aligned with the direction of the main scale of the particle, and the range difference of the point cloud projection coordinate on the three axes can respectively correspond to the long axis, the secondary long axis and the short axis of the particle; 2) For the volume V and the surface area S of the rock particle point cloud, constructing a surface grid model of the particle by adopting a Delaunay triangulation algorithm for calculation; for a given triangular mesh, the vertex position vector of any triangular surface patch is respectively Taking the origin as the fourth vertex of the tetrahedron, the total volume V of the entire mesh is the sum of all tetrahedron volumes: ; ; The surface area S can be obtained by summing the areas of all triangular patches: ; ; 3) Sphericity is used to measure how similar a rock particle shape is to an ideal sphere, which is defined as the ratio of the surface area of the particle to the surface area of a sphere of the same volume; ; Wherein V and S are distributed as the volume and surface area of the rock particles; 4) Roundness is used to measure the smoothness of the corners of a particle, wadell et al define three-dimensional roundness as the average of the ratio of the radius of curvature of each corner to the maximum inscribed sphere radius on the particle profile: ; Wherein R i is the sphere radius fitted by the ith edge angle, R is the radius of the largest inscribed sphere of the particle, N is the number of the particle edges; 5) The radius R 1 of the largest inscribed sphere of the particle and the minimum outer sphere R 2 of the particle adopt a Euclidean distance field calculation method, an internal point set P in is generated by uniformly sampling the inside of the particle, and for each internal sampling point, the minimum Euclidean distance and the maximum Euclidean distance between the internal sampling point and the outer surface of the particle are found, wherein the maximum value of the minimum distance corresponding to all the sampling points is the radius R 1 of the largest inscribed sphere, and the minimum value of the maximum distance corresponding to all the sampling points is the radius R 2 of the minimum outer sphere: ; 。
5. The three-dimensional point cloud based rockfill material grading detection method according to claim 1, wherein the step S5 comprises: S51, constructing a volume prediction model based on three-dimensional morphological characteristics of particles, taking three-dimensional geometric parameters such as a long axis, a minor axis, a short axis, a surface area, a volume and the like of the particles as input characteristics, and estimating the complete volume of the particles by adopting a machine learning regression method; S52, after the particle volume prediction is completed, setting particle size intervals according to engineering actual requirements by taking the secondary axis size of the particles as a particle size division basis, counting the proportion of the particle volume in each particle size interval, and drawing a discrete grading curve of the heap and the stones by taking the particle size as an abscissa and the volume distribution percentage as an ordinate.
6. The three-dimensional point cloud based rockfill material grading detection method according to claim 5, wherein the volume prediction model in S51 is a LightGBM regression model.

Description

Three-dimensional point cloud-based rock-fill material grading detection method Technical Field The invention belongs to the technical field of stone grading detection, and particularly relates to a three-dimensional point cloud-based stone grading detection method. Background The reservoir dam is an important infrastructure for realizing optimal allocation of water resources and guaranteeing flood control safety. The rock-fill dam has the advantages of local material availability, strong adaptability, good shock resistance, low carbon footprint, short construction period, low cost and the like, and becomes one of the most rapidly developed and competitive dam types. The stability and impermeability of the rock-fill dam mainly depend on the grading composition of the dam body filling materials, and reasonable grading is important for guaranteeing the integral quality and safe operation of hydraulic engineering. The traditional dam material grading detection method mainly adopts a screening method, has the problems of time and labor consumption, low efficiency, feedback lag and the like, and is difficult to meet the high standard requirements of the mechanized and intelligent construction of the modern earth-rock dam. With the development of three-dimensional laser scanning technology, a grading detection method based on point cloud is gradually raised. Compared with the traditional image recognition method, the point cloud data can more completely retain three-dimensional geometric information of particles, and the grading conversion precision is higher. However, the conventional point cloud segmentation method relies on manual design characteristics, is difficult to cope with complex scenes with dense particle stacking and remarkable scale difference, and has the problems of low segmentation precision, poor model generalization capability and the like. In recent years, the application of deep learning in the field of point cloud processing has been significantly advanced, and a segmentation algorithm based on the deep learning shows stronger recognition capability and adaptability in a complex environment. Therefore, it is needed to provide a method for detecting the grading of the rock-fill material based on deep learning and three-dimensional point cloud technology to solve the above problems. Disclosure of Invention Aiming at the defects in the prior art, the invention aims to provide a three-dimensional point cloud-based rock-fill grading detection method, which solves the problem of low segmentation precision of the traditional point cloud segmentation method. The technical scheme adopted by the invention is as follows: a three-dimensional point cloud-based rock-fill material grading detection method comprises the following steps: S1, collecting stone point cloud data; S2, projecting the rock-fill material point cloud along the overlooking direction to generate a corresponding depth image; S3, inputting the depth map into an example segmentation model to obtain a segmentation mask of each stone particle; S4, mapping the segmentation mask back to a three-dimensional point cloud, and extracting three-dimensional morphological characteristics of each particle; s5, predicting the volume and the particle size based on the three-dimensional characteristics of the particles, and drawing a grading curve. Further, the step S2 specifically includes the following steps: S21, rotating the stacking material point cloud to a standard horizontal coordinate system so as to be subjected to depth projection. Firstly, carrying out plane fitting on point clouds by adopting a random sampling consistency RANSAC algorithm, and extracting ground normal vectors from the point clouds. For any point P i, its distance to the fitting plane is:; When Di < epsilon (preset threshold), P i is regarded as the inner point of the current plane, and the RANSAC algorithm finally determines the ground normal vector by iteratively searching the plane with the largest inner point number 。 Then, the point cloud is rotated to a horizontal coordinate system by using a Rodrigues rotation formula, and a rotation matrix R 2 is: ; ; ; s22, projecting the rotated point cloud to an XOY plane along the Z axis, and constructing a regular two-dimensional grid. Setting the size of the grid cell to be delta x delta y, traversing all points, and determining the index position of the points in the grid according to X, Y coordinates of the points: ; In each grid unit, adopting a maximum height projection principle, and reserving the point with the maximum Z value in the area as a depth value so as to generate a corresponding depth matrix: ; The depth matrix D is converted into a gray level diagram form after normalization processing, a corresponding two-dimensional depth diagram is generated, and the gray level value of each pixel reflects the surface height information of the area and is used as the input of a subsequent example segmentation model. Further,