CN-121616634-B - Automatic point cloud extraction and registration method and device for underground tunneling roadway or tunnel

CN121616634BCN 121616634 BCN121616634 BCN 121616634BCN-121616634-B

Abstract

The application relates to an automatic point cloud extraction and registration method and device for underground tunneling tunnels or tunnels, which belong to the field of three-dimensional point cloud data and are technically characterized in that the point cloud of an original tunnel or tunnel is subjected to statistical filtering to remove noise, target point clouds are identified and extracted by utilizing the high reflection intensity characteristic of L-shaped plane targets and a plurality of L-shaped targets are segmented, each target point cloud is subjected to least square plane fitting, projected to a fitting plane and uniformly transformed to an XOY plane, two central axes of each target are extracted, the central axes are fitted by adopting a random sampling consistency algorithm and intersection points are solved, the coordinate of the target homonymous feature points is obtained by carrying out inverse transformation and restoring to an original coordinate system, coarse registration is carried out by utilizing at least four pairs of homonymous feature points through a quaternion method, and the problem of difficulty in estimating initial pose in various kinds of weak textures, large dust, strong water mist and other underground spaces is solved by combining an ICP algorithm, and the accuracy and the efficiency of point cloud registration are improved.

Inventors

XU SHAOYI
WANG HAO
WANG CHENGTAO
DENG XING
ZHU ZUHAO
YU CHENHUI
CHEN CHONG

Assignees

中国矿业大学

Dates

Publication Date: 20260508
Application Date: 20260203

Claims (10)

1. The automatic point cloud extraction and registration method for the underground tunneling roadway or tunnel is characterized by comprising the following steps of: Constructing a kd tree index structure in the original point cloud data and the target point cloud data, and performing statistical filtering to remove discrete noise points and reserve a tunnel or tunnel main body structure; Extracting target point clouds through dynamic intensity threshold values based on the high reflection intensity characteristics of the L-shaped plane targets, and performing European cluster segmentation on the extracted point clouds to identify and separate a plurality of L-shaped plane targets; performing least square plane fitting on each target point cloud respectively, projecting the target point cloud to a fitting plane, and unifying the target point cloud to an XOY plane through coordinate transformation so as to eliminate gesture differences; Extracting two orthogonal central axes of each L-shaped target based on an average coordinate statistics principle, adopting a random sampling consistency algorithm to perform central axis fitting, solving an intersection point, and restoring to an original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the same-name feature points of the targets; Calculating an initial transformation matrix by using at least four pairs of homonymous target feature points in the two-stage point cloud through a quaternion method for coarse registration, and then combining an ICP algorithm for fine registration to realize high-precision alignment of the point cloud; The method for obtaining the three-dimensional coordinates of the target homonymous feature points comprises the following steps: extracting homonymous characteristic points by adopting a statistical analysis algorithm, randomly generating a plurality of random intervals in a given range, determining the points of all point clouds in each interval, of which the coordinates are located in the interval, respectively calculating a coordinate average value for each interval, and generating a new point cloud at the coordinates of the coordinate average value; Performing straight line fitting on the new point cloud, repeating the process for a plurality of times, selecting a model with the most inner points as a final estimation result, and fitting out approximate axes of two long sides of the L-shaped plane target; And constructing a matrix, applying a linear regression model to calculate the slope and intercept of two straight lines, calculating the coordinates of the intersection point through an intersection point formula of a straight line equation, and obtaining the coordinates of the homonymous characteristic points of the L-shaped plane target through inverse transformation of rotation and translation of the fitting plane.
2. The method for automatically extracting and registering point clouds of an underground tunneling roadway or tunnel according to claim 1, wherein the method further comprises the steps of acquiring original point cloud data before the L-shaped plane targets are identified and extracted, preprocessing the original point cloud data to obtain roadway or tunnel point clouds, wherein the preprocessing comprises the steps of manually cutting out the original point cloud data, removing the incompletely scanned partial roadway or tunnel point clouds and a large-range noise point cloud, and removing individual discrete point clouds by using a statistical filtering denoising method.
3. The method for automatically extracting and registering point clouds of an underground tunneling roadway or tunnel according to claim 1, wherein the step of constructing kd-tree index structure in original point cloud data and target point cloud data, performing statistical filtering to remove discrete noise points and preserve roadway or tunnel main body structure comprises: For the acquired original point cloud data and target point cloud data, each layer of the kd-tree uses a hyperplane perpendicular to a corresponding axis to divide all the sub-levels along a preset dimension, wherein the sub-levels are divided along the preset dimension, particularly at the root of the tree, all the sub-nodes are divided according to a first dimension, each layer in the tree is divided in a next dimension, and when other dimensions are exhausted, the tree returns to the first dimension; Filling point cloud data into a kd tree data structure, accelerating statistical analysis of all points in the point cloud, and calculating average distance between the points and k adjacent points: Wherein, the Is a dot The distance from a point in its neighborhood, Is a dot The average distance from the point in its neighborhood, Is a dot Is provided with a plurality of three-dimensional coordinates, Is that Any point coordinates within the KNN neighborhood of (c), Is the number of neighbor points, namely the K value in KNN; Calculating the average distance and standard deviation of global points in the point cloud: setting statistical filtering threshold Wherein Representing the average value of the average distances from all points in the point cloud to the KNN neighborhood of the points; It is the coefficient that is calculated and, Is the standard deviation of the average distance from all points in the point cloud to the neighborhood thereof; is the total number of points in the point cloud, when > And traversing all points in the point cloud to complete the statistical filtering of the point cloud.
4. The method for automatically extracting and registering point clouds of an underground tunneling roadway or tunnel according to claim 1, wherein the step of extracting target clouds by a dynamic intensity threshold based on the high reflection intensity characteristic of the L-shaped planar targets and performing euclidean clustering segmentation on the extracted point clouds to identify and separate a plurality of L-shaped planar targets comprises: Selecting a target point cloud region to perform rough extraction, and calculating the average value of the maximum and minimum return reflection intensities in the target point cloud region obtained by rough extraction Will be 1.5 times As a threshold value, extracting point clouds larger than the threshold value as target point cloud fine extraction results; Selecting a seed point as a starting point of a current cluster, traversing all unclassified points, calculating Euclidean distance between the unclassified points and the seed point, and classifying the points into the same cluster as the seed point if the distance is smaller than a preset threshold value; the above procedure is repeated for points in the same cluster, and points having a distance from any point in the current cluster less than the threshold value are added to the cluster until all points are classified into a certain cluster.
5. The automatic extraction and registration method of point clouds of underground tunneling tunnels or tunnels according to claim 1 is characterized in that the method comprises the steps of performing least square plane fitting on each target point cloud, projecting the target point clouds to a fitting plane, unifying the coordinate transformation to an XOY plane to eliminate posture differences, and obtaining a rotated point cloud by projecting three-dimensional point clouds onto the best fitting plane, converting the projected coordinates of the point clouds and rotating coordinates to obtain a projection plane of an L-shaped target, obtaining a transformation matrix by utilizing a Rodrigues formula according to a normal vector of the plane, and multiplying the coordinate of the target point clouds on the fitting plane by a rotation matrix R.
6. The automatic point cloud extraction and registration method for underground tunneling tunnel or tunnel according to claim 1, wherein the step of performing coarse registration by calculating an initial transformation matrix by quaternion method and performing fine registration by combining ICP algorithm by utilizing at least four pairs of homonymous target characteristic points in two-stage point clouds to realize high-precision alignment of point clouds comprises extracting four pairs of homonymous coordinate points of two-stage point clouds by utilizing L-target homonymous characteristic point extraction method, solving transformation matrix between original point cloud data and target point cloud data according to SVD matrix decomposition method, And after the integral transformation of the original point cloud data, the original point cloud data is spliced with the target point cloud data to finish coarse registration, and then ICP registration is performed on the two-stage point cloud to finish fine registration of the point cloud.
7. An automatic point cloud extraction and registration device for an underground tunneling roadway or tunnel is characterized by comprising: The acquisition unit is configured to construct a kd tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and reserve a main structure of a tunnel or a tunnel; The first processing unit is configured to extract target point clouds through dynamic intensity threshold values based on the high reflection intensity characteristics of the L-shaped plane targets, and perform European cluster segmentation on the extracted point clouds so as to identify and separate a plurality of L-shaped plane targets; the second processing unit is used for carrying out least square plane fitting on each target point cloud respectively, projecting the target point clouds to a fitting plane, and unifying the target point clouds to an XOY plane through coordinate transformation so as to eliminate gesture differences; The third processing unit is configured to extract two orthogonal central axes of each L-shaped target based on an average coordinate statistics principle, perform central axis fitting by adopting a random sampling consistency algorithm, calculate an intersection point, restore to an original coordinate system through inverse transformation, and obtain three-dimensional coordinates of the same-name feature points of the targets; And the determining unit is configured to perform quaternion coarse registration by utilizing the coordinates of the characteristic points of the two-stage point cloud targets, and then perform fine registration by adopting an ICP algorithm.
8. The automatic point cloud extraction and registration device for an underground tunneling roadway or tunnel according to claim 7, which is suitable for a coal cutter, a heading machine, a shield machine or tunnel mining equipment of an underground space.
9. A computer readable storage medium having stored thereon a computer program, which when executed by a computing device causes the computing device to perform the method of point cloud automatic extraction and registration of an underground excavation roadway or tunnel of any one of claims 1 to 6.
10. An electronic device comprising one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising a point cloud automatic extraction and registration method for performing an underground roadway or tunnel of any one of claims 1-6.

Description

Automatic point cloud extraction and registration method and device for underground tunneling roadway or tunnel Technical Field The application relates to the technical field of three-dimensional point cloud data, in particular to a method and a device for automatically extracting and registering point cloud of an underground tunneling roadway or tunnel. Background Along with the development of technology and the improvement of industrial automation level, intelligent construction of coal mine tunnels and tunnel tunneling has become a key for improving coal mine safety and efficiency. In the intelligent construction of coal mine tunnel and tunnel tunneling, the construction of an accurate three-dimensional model of the tunnel or tunnel is a basis for realizing the application of surrounding rock deformation monitoring, roof bolt classification and the like of the tunnel or tunnel. To achieve this goal, synchronous localization and mapping (Simultaneous Localization AND MAPPING, SLAM for short) technology is the core of research. The laser SLAM technology is not influenced by dust shielding and illumination change, so that accurate three-dimensional space distance information can be obtained in real time, and the laser SLAM technology is particularly suitable for environments with low illumination and high dust such as underground coal mines and underground tunnel spaces. In the laser SLAM technology, point cloud registration is a key process, and involves spatial alignment of point cloud data acquired at different times or different positions to ensure continuity and consistency of mapping, and the point cloud registration is generally divided into two stages of coarse registration and fine registration. The purpose of the coarse registration is to estimate the approximate transformation relationship between the two point clouds, and provide initial estimation for fine registration for realizing high-precision point cloud alignment. At present, researchers propose various point cloud coarse registration algorithms, including a registration method based on geometric feature description and a registration method based on global search ideas. Although these methods improve the accuracy and efficiency of registration to some extent, certain limitations remain in this particular environment of coal mine undermining or underground tunnelling. For example, although the PCA algorithm based on geometric feature description can calculate covariance matrix and rotation matrix of multi-site cloud, but has lower robustness to noise and outlier, and the RANSAC algorithm based on global search idea can process data containing noise, but has lower algorithm stability due to single constraint condition based on probability, and is easy to generate mismatching. Therefore, a plane target-assisted generalized four-point consistent rough registration algorithm is adopted. However, the existing target feature point extraction method is not high in applicability when processing a single long and narrow structure such as a coal mine, a roadway and a tunnel in an underground space, and in practical application, due to the influence of factors such as low light, humidity and noise, the phenomenon of missing or redundancy of the planar target point cloud data obtained by scanning is easy to occur, so that the difficulty of point cloud registration is further increased. Therefore, developing a new target point cloud feature point extraction and registration method suitable for underground space mining to improve the accuracy and efficiency of point cloud registration becomes a technical problem to be solved urgently. Disclosure of Invention The embodiments of the application provide a method and a device for automatically extracting and registering point clouds of an underground tunneling roadway or tunnel, so as to at least have the technical effect of improving the accuracy and the efficiency of point cloud registration in a low-light high-dust environment. In order to solve the above technical problems, according to an aspect of the present application, there is provided a method for automatically extracting and registering point clouds of an underground tunnel or tunnel, including: constructing kd tree index structures in the original point cloud data and the target point cloud data, and performing statistical filtering to remove discrete noise points and reserve a main structure of a tunnel or a tunnel; Extracting target point clouds through dynamic intensity threshold values based on the high reflection intensity characteristics of the L-shaped plane targets, and performing European cluster segmentation on the extracted point clouds to identify and separate a plurality of L-shaped plane targets; performing least square plane fitting on each target point cloud respectively, projecting the target point cloud to a fitting plane, and unifying the target point cloud to an XOY plane through coordinate transformation so as to eliminate