CN-121999022-A - Tunnel point cloud registration method based on slice contour map matching

CN121999022ACN 121999022 ACN121999022 ACN 121999022ACN-121999022-A

Abstract

The invention relates to the technical field of three-dimensional point cloud data processing and registration, discloses a tunnel point cloud registration method based on slice contour map matching, and solves the problems of low registration accuracy, insufficient robustness and poor adaptability of the existing tunnel point cloud registration method under the conditions of low overlapping rate, strong noise and structural repetition. According to the scheme, the tunnel point cloud is preprocessed, noise points are removed, the coordinate scale is unified, the main axis direction of the tunnel is extracted through principal component analysis, and curvature-driven self-adaptive slicing is performed along the main direction. And extracting the contour by using ALPHA SHAPE algorithm for each slice, generating a uniform point sequence by polar angle sequencing and resampling, and obtaining a smooth curve by Catmull-Rom spline fitting. And then, converting the slice contour into a graph structure, and carrying out node matching and transformation estimation through graph similarity optimization to realize high-precision registration of the multi-site tunnel point cloud.

Inventors

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

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. The tunnel point cloud registration method based on slice contour map matching is characterized by comprising the following steps of: s1, preprocessing a source point cloud and a target point cloud, and determining the direction of a tunnel main shaft through principal component analysis; S2, slicing the source point cloud and the target point cloud along the direction of the tunnel main shaft by adaptively adjusting the slicing distance to obtain slice point clouds; s3, extracting contours from each slice point cloud; S4, converting the contour of each slice into a contour map structure containing node sets, edge sets and characteristic information; s5, establishing a contour map node matching relationship between the source point cloud and the target point cloud based on a map matching algorithm; s6, estimating rigid transformation parameters from the source point cloud to the target point cloud according to the node matching relationship; S7, registering the source point cloud and the target point cloud by applying the rigid transformation parameters.
2. The method for registration of tunnel point clouds based on slice contour map matching as claimed in claim 1, In the step S1, the preprocessing comprises the steps of filtering out isolated noise points by adopting a statistical outlier removal algorithm, and carrying out coordinate centering and scale normalization processing on point cloud.
3. The method for registration of tunnel point clouds based on slice contour map matching as claimed in claim 1, In the step S2, the slice interval is adaptively adjusted based on curvature change rate, point density distribution or information entropy; the method for adaptively adjusting the slice spacing based on the curvature change rate comprises the following steps: Calculating a local curvature for each point The definition is: ; Wherein, the Is taken as a point Eigenvalues of the neighborhood covariance matrix; let the slice spacing along the tunnel main axis direction be The following steps are: ; Wherein, the Is the base slice spacing; is the global average curvature; Is the maximum curvature; is an adaptive coefficient.
4. The method for registration of tunnel point clouds based on slice contour map matching as claimed in claim 3, In step S2, slicing the source point cloud and the target point cloud to obtain a slice point cloud, including: Along the main axis Direction, as the starting point For reference, sequentially generating slice planes: ; Wherein, the Represent the first A slice plane; Representing a point coordinate vector; Represent the first The spacing between the individual slice and the previous slice; in each slice plane intercept point cloud: ; Is defined as the ith slice point cloud ; Wherein, the Representing the first point in the first point cloud A plurality of points; is a preset slice thickness threshold.
5. The method for registration of tunnel point clouds based on slice contour map matching as claimed in claim 1, In step S3, extracting the contour for each slice point cloud includes: For each slice point cloud, taking the tunnel main axis direction as a normal direction, projecting the point set onto a two-dimensional plane to obtain a projection point set; For the projection point set, extracting a slice boundary point set by adopting polar angle sorting or ALPHA SHAPE algorithm; performing sparse repair and abnormal point rejection on the slice boundary point set; And (3) performing curve fitting by adopting Catmull-Rom spline or B spline interpolation to generate a smooth and continuous contour curve.
6. The method for registration of tunnel point clouds based on slice contour map matching as claimed in claim 1, In step S4, converting the contour of each slice into a contour map structure including node sets, edge sets and feature information, including: s41 contour curves for each slice Resampling it to n equidistant points to obtain a uniform point set , , wherein, Representing the first of the uniform point set Sampling points; s42, regarding each sampling point in the uniform point set as a node in the profile graph, and defining a feature vector: ; Wherein, the Is a node Curvature at; Is a node Polar angle relative to the centroid of the profile; Is a node Radial distance from the centroid of the profile; For from the start point to the node of the profile Arc length of (2) to represent node A position on the contour; s43, normalizing the node characteristics of the graph: ; Wherein, the Is the maximum of the curvatures of all nodes in the profile; is the maximum value in the radial distance between all nodes in the contour and the contour centroid; Is the total arc length of the outline; s44, establishing edges between adjacent nodes on the outline, wherein the weight of the edges is defined as: ; Wherein, the Representing nodes And node Euclidean distance between them; Representing nodes Curvature at; Representing nodes Curvature at; Representing the curvature weight coefficient.
7. The method for matching tunnel point cloud registration based on slice contour map as recited in claim 6, wherein, In step S5, establishing a profile node matching relationship between the source point cloud and the target point cloud based on a graph matching algorithm, including: S51, respectively extracting characteristic information in a contour map of a source point cloud and a contour map of a target point cloud, wherein the characteristic information comprises node characteristics and side weight characteristics; S52, based on similarity between node characteristics and structural consistency between edge weight characteristics of a contour map of a source point cloud and a contour map of a target point cloud, constructing a node matching cost matrix through weighted fusion: ; Wherein, the Nodes in a profile for a source point cloud Nodes in a profile with a target point cloud Is matched with the total cost; Is a node Is included in the normalized feature vector of (a); Is a node Is included in the normalized feature vector of (a); Represents the L2 paradigm; Is a side weight difference weight coefficient; Nodes in a profile for a source point cloud With adjacent nodes Is a side weight of (2); Is a node in the outline map of the target point cloud With adjacent nodes Is a side weight of (2); S53, solving a global optimal node matching relation by adopting an optimization algorithm based on the node matching cost matrix, and converting the matching problem into binary matching for minimizing the total cost during solving: ; Wherein, the Nodes in a contour map representing a source point cloud Nodes in a profile with a target point cloud If equal to 1, it indicates a match, and if equal to 0, it indicates a mismatch.
8. The method for matching tunnel point cloud registration based on slice contour map as recited in claim 7, Step S53 further includes: introducing global topological consistency constraint when solving the global optimal node matching relation: Defining topology consistency errors among the matched nodes as follows: ; Wherein, the Is a global topology consistency error; index pairs for nodes in the profile of the source point cloud, ; Index pairs for nodes in the profile of the target point cloud, ; Nodes in a contour map representing a source point cloud Nodes in a profile with a target point cloud Is a match of the matching states of (a); Nodes in a contour map representing a source point cloud Nodes in a profile with a target point cloud Is a match of the matching states of (a); The final matching result is obtained by the following joint optimization: ; Wherein, the As a characteristic error term, Is a weight coefficient.
9. The method for matching tunnel point cloud registration based on slice contour map as recited in claim 8, Step S5 further includes: S54, fusing the matching results of the multiple slices: Calculating the average matching cost of the corresponding slice according to the matching cost of each node in the contour map corresponding to the slice; based on the slice average matching cost, defining a matching confidence of each slice: ; Wherein, the Is the first point in the source point cloud Confidence of matching of individual slices; Is the first point in the source point cloud Average matching cost of individual slices; Standard deviation of average matching cost of all slices in the source point cloud; confidence of matching point pairs for all slices in a source point cloud according to matching of corresponding slices And weighting to obtain a globally uniform matching point pair set, namely obtaining the profile graph node matching relationship of the final source point cloud and the target point cloud.
10. The method for registration of tunnel point clouds based on slice contour map matching according to any one of claims 1 to 9, wherein in step S6, estimating the rigid transformation parameters from the source point cloud to the target point cloud according to the node matching relationship includes: s61, solving initial rigidity transformation parameters through a least square method based on the node matching relation; S62, optimizing by adopting a RANSAC algorithm based on the initial rigid transformation parameters to obtain steady rigid transformation parameters; S63, based on the slice matching confidence, weighting and fusing the steady transformation parameters of all slices to obtain global rigid transformation parameters , wherein, In the form of a global rotation matrix, Is a global translation vector.

Description

Tunnel point cloud registration method based on slice contour map matching Technical Field The invention relates to the technical field of three-dimensional point cloud data processing and registration, in particular to a tunnel point cloud registration method based on slice contour map matching. Background With the rapid development of space sensing technologies such as three-dimensional laser scanning (LiDAR), instant positioning and map building (SLAM), tunnel point cloud data are widely applied to engineering construction, safety monitoring and digital twin modeling. In general, tunnel point cloud data is acquired through multi-station segmented scanning, and due to inherent deviation of coordinate systems of different stations, the scanning process is easily interfered by environment, and spatial alignment of multi-station data needs to be achieved through registration. The existing main stream point cloud registration method mainly comprises three types: (1) The geometrical registration method based on the closest point of iteration relies on point cloud data with high overlapping rate and a good initial alignment state, and when the overlapping rate of the tunnel point cloud is low, noise interference is strong or the structure is repeated, local optimal solution is easy to fall in, so that registration failure or accuracy is greatly reduced. (2) The local matching method based on the feature descriptors is difficult to extract stable key points and descriptors due to single geometric texture of the tunnel wall surface, has insufficient matching robustness, and can not meet the engineering level precision requirement. (3) The end-to-end registration method based on deep learning has potential in partial scenes, but depends on large-scale annotation data set training, has high requirement on hardware computing power, and has poor adaptability in complex field environments of tunnel engineering. In addition, tunnel point clouds generally have large-scale, long-distance extension, non-rigid noise accumulation, etc., further increasing the difficulty of robust registration. Therefore, a tunnel point cloud processing method capable of realizing high-precision and stable registration under the conditions of low overlapping rate, strong noise and structural repetition is needed to improve the precision and the robustness of three-dimensional reconstruction and deformation monitoring of a tunnel. Disclosure of Invention The invention aims to solve the technical problems of low registration precision, insufficient robustness and poor adaptability of the existing tunnel point cloud registration method under the conditions of low overlapping rate, strong noise and structural repetition. The technical scheme adopted for solving the technical problems is as follows: A tunnel point cloud registration method based on slice contour map matching comprises the following steps: s1, preprocessing a source point cloud and a target point cloud, and determining the direction of a tunnel main shaft through principal component analysis; S2, slicing the source point cloud and the target point cloud along the direction of the tunnel main shaft by adaptively adjusting the slicing distance to obtain slice point clouds; s3, extracting contours from each slice point cloud; S4, converting the contour of each slice into a contour map structure containing node sets, edge sets and characteristic information; s5, establishing a contour map node matching relationship between the source point cloud and the target point cloud based on a map matching algorithm; s6, estimating rigid transformation parameters from the source point cloud to the target point cloud according to the node matching relationship; S7, registering the source point cloud and the target point cloud by applying the rigid transformation parameters. Further, in step S1, the preprocessing comprises filtering out isolated noise points by adopting a statistical outlier removal algorithm, and carrying out coordinate centering and scale normalization processing on point cloud. Further, in step S2, the slice interval is adaptively adjusted based on the curvature change rate, the dot density distribution or the information entropy; the method for adaptively adjusting the slice spacing based on the curvature change rate comprises the following steps: Calculating a local curvature for each point The definition is: ; Wherein, the Is taken as a pointEigenvalues of the neighborhood covariance matrix; let the slice spacing along the tunnel main axis direction be The following steps are: ; Wherein, the Is the base slice spacing; is the global average curvature; Is the maximum curvature; is an adaptive coefficient. Further, in step S2, slicing the source point cloud and the target point cloud to obtain a slice point cloud, including: Along the main axis Direction, as the starting pointFor reference, sequentially generating slice planes: ; Wherein, the Represent the firstA slice pla