Tunneling roadway point cloud registration method based on 3D NDT-ICP algorithm

Abstract

The invention provides a tunneling roadway point cloud registration method based on a 3DNDT-ICP algorithm, which is used for improving the environment modeling precision of a tunneling roadway and meeting the intelligent perception requirement of the tunneling environment. The method comprises the following steps: firstly, preprocessing tunneling roadway point clouds through a VoxelGrid filtering method, and reducing the number of the point clouds while maintaining the overall structure of the point clouds; then performing coordinate transformation solution on the tunneling roadway point cloud by using a 3DNDT algorithm, carrying out parameter optimization on algorithm cell resolution by combining tunneling roadway environment characteristics, sending the coordinate transformation parameters obtained by solution to an ICP algorithm, and initializing a point cloud seven-parameter coordinate matrix in the ICP algorithm; and finally, introducing a KD-Tree in an ICP algorithm to perform point pair search, and optimizing algorithm nonlinear objective function solution by adopting a Gauss-Newton method to complete tunneling roadway point cloud accurate registration.

Description

technical field [0001] The invention relates to a method for point cloud registration of an excavation roadway under a mine, which is used for improving the modeling progress of the excavation roadway environment and meeting the demand for intelligent perception of the excavation roadway. Background technique [0002] In recent years, the low efficiency of excavation in our country has led to unbalanced excavation, and the problem of continuous excavation has become the main problem faced by high-efficiency mines. Due to the complex underground environment, it is difficult for excavation equipment to perceive the environment, and it is difficult to restrict its own movement space and ensure the quality of roadway formation. Therefore, it is urgent to improve the intelligence level of the excavation working face and realize the intelligent perception of the excavation roadway environment as soon as possible. As the excavation roadway is an independent confined space undergrou...

AI Technical Summary

Problems solved by technology

This patent can improve the accuracy of the algorithm, but the method has a large amount of calculation, and there are still some shortcomings in improving efficiency

[0004] The patent number is CN111161327A, and the authorized publication date is 2020.05.15. The invention patent "a point cloud registration method combining rotating platform and ICP" discloses a point cloud registration method combining rotating platform and ICP Through the rough registration of the rotating platform, it does not need to rely on the surface features of the point cloud itself, but only needs to know the rotation angle and the position of the rotation axis between the point clouds, and then use this registration result as the initial position of the ICP registration algorithm for further processing Fine registration, which effectively solves the problem of point cloud registration failure due to similar or even identical features at various angles. This patent simplifies the operation steps of point cloud registration and makes the experiment more convenient, but there is still low matching accuracy. And the problem of low efficiency, the final iterative result falls into a local optimal situation

[0005] The patent number is CN112017219A, and the authorized publication date is 2020.12.01. The invention patent "a laser point cloud registration method" provides a laser point cloud registration method. After the 3D point cloud and the target 3D point cloud are respectively down-sampled and segmented into non-ground point clouds, the source local features and target local features corresponding to the local features of the source 3D point cloud and the target 3D point cloud are extracted and feature matched, and then based on The matching result estimates the pose transformation of the source 3D point cloud relative to the target 3D point cloud, and obtains the pose transformation matrix between the source 3D point cloud and the target 3D point cloud to complete the matching of the source 3D point cloud and the target 3D point cloud Registration, this method can reduce the calculation pressure while retaining the information of the original point cloud through downsampling, but it will cause errors when extracting the local features of the source 3D point cloud and the target 3D point cloud. , there is a defect of low precision

Images