CN-121999021-A - Rock mass point cloud registration method based on convex-concave characteristics

CN121999021ACN 121999021 ACN121999021 ACN 121999021ACN-121999021-A

Abstract

The invention relates to the field of computer vision and image processing, discloses a rock mass point cloud registration method based on convex-concave features, and solves the problems of unstable feature extraction, weak descriptor distinguishing capability, insufficient registration precision and robustness of the existing rock mass point cloud registration method under a complex scene. According to the scheme, an improved RNRLC algorithm is utilized to extract concave-convex characteristic points of a rock surface layer, stability of key points under the conditions of noise and uneven density is guaranteed, then a local coordinate system is established by taking each concave-convex characteristic point as the center, a neighborhood direction point signature histogram descriptor is calculated, concave-convex weights are introduced in an interpolation voting process, sensitivity and stability of the descriptor to geometric structure differences are enhanced, matching confidence is comprehensively evaluated according to Euclidean distance and cosine similarity of descriptor vectors, an initial corresponding point pair is screened out, finally, an RANSAC algorithm is adopted to reject mismatching and evaluate a rigid transformation matrix, and the method is applied to rock surface layer source point cloud data, and high-precision registration is achieved.

Inventors

XIAO HUABO
SHI WEIMING
LIU SHIYONG
ZHANG HAN
LIN HAODONG
XIAO FENG

Assignees

中国电建集团成都勘测设计研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. A rock mass point cloud registration method based on convex-concave characteristics, which is characterized by comprising the following steps: s1, inputting rock mass point cloud data to be registered, wherein the rock mass point cloud data comprises a source point cloud and a target point cloud; S2, calculating the geometric center of the point cloud, constructing a kd tree and calculating the normal vector of the point cloud; s3, identifying concave-convex feature points of the rock mass by analyzing the normal vector of the point cloud, and filtering noise points by a clustering algorithm; S4, calculating the mass center of the clustering result, and constructing a local coordinate system based on the mass center; S5, constructing a concave-convex characteristic weighted descriptor according to the local coordinate system, and integrating concave-convex information in the descriptor generation process; S6, calculating the Euclidean distance between the concave-convex characteristic points of each of the two point clouds to be matched, screening point pairs meeting a threshold value, and constructing an initial matching point pair set; s7, eliminating error matching point pairs aiming at the initial matching point pair set, and calculating a rigid transformation matrix; s8, applying the rigid transformation matrix to the source point cloud to realize the registration of the two rock mass point clouds.
2. A method for rock mass point cloud registration based on convex-concave features as defined in claim 1, In the step S3, analyzing the normal vector of the point cloud through an R-NRLC algorithm, and identifying concave-convex characteristic points of the rock mass: computing a local neighborhood for each point Centroid of (2) Centroid is the current point Coordinates and feature vectors of (a) And (3) summing; Feature vector Normal vector along the current point Decomposition into vertical components And horizontal component For each intra-domain neighboring point of the point Vectors formed Respectively decomposing along the vertical and horizontal directions to obtain two components And ; Will be the vertical component And (3) with Adding up the horizontal component And (3) with Adding to obtain new feature vector, defined as ; According to Normal vector Dot product result of (2) Distinguishing the region attribute of the feature point, wherein the dot product result is that the regular point is positioned in the convex region, and the dot product result is that the dot product is negative and the dot product is positioned in the concave region; According to Normal vector Positive and negative dot product values of (2) will be local neighborhoods Dividing the neighborhood points in the neighborhood region into two groups, counting the proportion of the two groups of points to the total number of the local neighborhood points, and marking the two groups of points as a and b; the weights of the two groups of points are calculated through a sigmoid function, and the sigmoid function is expressed as: ; Wherein, the To control sensitivity of noise suppression effects; Definition of the definition , Will be And Respectively used as the weight of the adjacent points in the corresponding group, and the current point is calculated by weighting Sum of feature vectors to all neighboring points ; Setting a direction threshold And a length threshold , wherein, Is used for controlling the direction screening of the concave-convex area of the point cloud, For controlling relief areas is selected according to the degree of variation of the (a); Calculation of Normal vector Dot product absolute value of (2) Will be Greater than Marking the point of interest as the point of interest, and screening the characteristic vector from the point of interest Less than As the final convex-concave feature points.
3. A method for rock mass point cloud registration based on convex-concave features as defined in claim 1, In step S3, the clustering algorithm adopts DBSCA algorithm.
4. A method for rock mass point cloud registration based on convex-concave features as defined in claim 1, In step S4, a centroid of the clustering result is calculated, and a local coordinate system is constructed based on the centroid, including: Calculating the geometric center of each point cloud cluster obtained through clustering, and taking the geometric center as a stable centroid; Performing principal component analysis on each point cloud cluster, calculating a covariance matrix of the point cloud cluster, and decomposing the feature values to obtain three feature vectors 、、 The directions of the maximum eigenvalue, the medium eigenvalue and the minimum eigenvalue of the covariance matrix are respectively corresponding; Feature vector with stable centroid as origin 、、 And respectively constructing a three-dimensional orthogonal local coordinate system in the Z-axis direction, the Y-axis direction and the X-axis direction of the local coordinate system.
5. A method for rock mass point cloud registration based on convex-concave features as defined in claim 1, In step S5, constructing a concave-convex feature weighted descriptor according to the local coordinate system, including: In a local coordinate system, dividing a plurality of spherical subregions by taking a target point as a center, and projecting a neighborhood point of the target point to the corresponding spherical subregion according to a space position; Extracting direction distribution information from the neighborhood points in each spherical subarea according to the included angle relation between the normal vector and the coordinate axis of the local coordinate system; Defining a concave-convex weight factor, wherein the concave-convex weight factor is defined by the concave-convex index of the point And distance weight Together, the determination, wherein, The convex point is indicated as being a convex point, Representing pits; In the direction information interpolation process, concave-convex weight factors are integrated into direction distribution statistics, and the direction information in each spherical subarea is weighted and accumulated to generate a weighted direction histogram of the subarea; And splicing the weighted direction histograms of all the spherical subregions according to a preset sequence to form a point signature descriptor containing convex-concave characteristic information.
6. A method for rock mass point cloud registration based on convex-concave features as defined in claim 5, In step S6, screening the point pairs meeting the threshold includes: calculating Euclidean distance and cosine similarity between the weighted descriptors of the source point cloud concave-convex characteristic points and the weighted descriptors of the target point cloud concave-convex characteristic points; Setting a Euclidean distance threshold and a cosine similarity threshold, and screening point pairs which simultaneously meet the Euclidean distance smaller than the Euclidean distance threshold and the cosine similarity larger than the cosine similarity threshold; and sorting the screened point pairs according to cosine similarity from high to low, removing the point pairs which do not meet geometric consistency, and finally forming an initial matching point pair set.
7. A method for rock mass point cloud registration based on convex-concave features as defined in claim 6, In step S7, for the initial matching point pair set, a RANSAC algorithm is adopted to reject the mismatching point pair, and a rigid transformation matrix is calculated, which specifically includes: S71, randomly selecting non-coplanar matching point pairs from the initial matching point pair set, and calculating a temporary rigid transformation matrix based on an SVD algorithm; S72, applying the tentative rigid transformation matrix to the source point cloud characteristic points, and calculating Euclidean distance errors between the transformed source points and corresponding target points; s73, marking point pairs with Euclidean distance errors smaller than a set distance error threshold as interior points, and counting the number of the interior points; S74, repeating the steps S71-S73, and after iteration setting times, selecting a tentative rigid transformation matrix obtained in one iteration with the largest number of inner points as an estimated optimal rigid transformation matrix.
8. A method for rock mass point cloud registration based on convex-concave features as defined in claim 7, In step S8, applying the rigid transformation matrix to the source point cloud includes: ; Wherein, the The registered source point cloud coordinates are obtained; The original source point cloud coordinates; is a rotation parameter in the rigid transformation matrix; is a translation parameter in the rigid transformation matrix.
9. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the method for rock mass point cloud registration based on convex-concave features as claimed in any one of claims 1-8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program when executed by a processor implements the method for rock mass point cloud registration based on convex-concave features as claimed in any one of claims 1 to 8.

Description

Rock mass point cloud registration method based on convex-concave characteristics Technical Field The invention relates to the field of computer vision and image processing, in particular to a rock mass point cloud registration method based on convex-concave characteristics. Background In the fields of rock mass structure analysis, geological disaster monitoring, engineering mapping, digital twin reconstruction and the like, high-precision three-dimensional point cloud data is an important foundation for realizing accurate analysis and modeling. Compared with the traditional two-dimensional image, the three-dimensional point cloud can more intuitively reflect the space geometric features and the structural morphology of the rock mass surface layer. In recent years, the development of lidar and three-dimensional reconstruction techniques has enabled high-precision scanning of rock surfaces. However, the method is limited by factors such as complex terrain shielding, observation view angle limitation and the like, the three-dimensional structure of the rock mass is difficult to be completely obtained by single scanning, and a complete rock mass digital model can be built only by registration fusion of multi-view and multi-batch point cloud data. However, rock mass point cloud registration faces a plurality of technical bottlenecks that a rock mass surface concave-convex structure is complex, natural textures are scarce, and the conventional registration method is difficult to extract stable and reliable features due to the fact that the rock mass surface concave-convex structure is easily influenced by factors such as environmental noise and uneven point cloud density in a scanning process, the geometrical registration method based on iteration closest points depends on high overlapping rate and good initial alignment, is easy to sink into a local optimal solution in a low overlapping rate and strong noise scene, the registration method based on conventional feature descriptors is insufficient in robustness when distinguishing similar geometric forms due to lack of targeted depiction of the rock mass concave-convex structure, and the deep learning registration method depends on a large-scale labeling data set, has high requirements on hardware calculation force and is difficult to adapt to complex application scenes of a construction site. In the prior art, part of methods try to construct an optimized registration effect through feature point extraction and descriptors, but the problems of poor stability of the feature points, insufficient sensitivity of the descriptors to concave-convex structures and the like generally exist, and in rock mass point cloud registration with low overlapping rate, high deflection angle and uneven density, the registration precision and robustness are difficult to be simultaneously achieved. Therefore, a high-reliability rock mass point cloud registration technology capable of accurately capturing concave-convex characteristics of a rock mass, effectively inhibiting noise interference and adapting to a complex scene is needed to break through the application limitation of the existing method. Disclosure of Invention The technical problem to be solved by the invention is to provide a rock mass point cloud registration method based on convex-concave characteristics, which solves the problems of unstable feature extraction, weak descriptor distinguishing capability and insufficient registration precision and robustness of the existing rock mass point cloud registration method under a complex scene. The technical scheme adopted for solving the technical problems is as follows: in a first aspect, the invention provides a rock mass point cloud registration method based on convex-concave characteristics, which comprises the following steps: s1, inputting rock mass point cloud data to be registered, wherein the rock mass point cloud data comprises a source point cloud and a target point cloud; S2, calculating the geometric center of the point cloud, constructing a kd tree and calculating the normal vector of the point cloud; s3, identifying concave-convex feature points of the rock mass by analyzing the normal vector of the point cloud, and filtering noise points by a clustering algorithm; S4, calculating the mass center of the clustering result, and constructing a local coordinate system based on the mass center; S5, constructing a concave-convex characteristic weighted descriptor according to the local coordinate system, and integrating concave-convex information in the descriptor generation process; S6, calculating the Euclidean distance between the concave-convex characteristic points of each of the two point clouds to be matched, screening point pairs meeting a threshold value, and constructing an initial matching point pair set; s7, eliminating error matching point pairs aiming at the initial matching point pair set, and calculating a rigid transformation matrix; s8, applying the