A point cloud registration method based on non-local operation

Abstract

The invention discloses a point cloud registration method based on non-local operations. First, design a non-local operation that can capture the information of the non-local region of the point cloud, and enhance the structural information of the point cloud itself; at the same time, design a cross non-local operation that can handle two different point clouds, and enhance the key matching information between the two point clouds to be matched ; Then design a deep learning network that processes two point clouds to be matched at the same time, and find the matching relationship between the point clouds by generating virtual corresponding points; finally use the singular value decomposition to obtain the rotation and transformation between the two point clouds to be matched The translation transformation matrix. The point cloud registration method based on non-local operations proposed by the present invention has strong robustness to noise, outlier points and point cloud inputs of different densities, can effectively prevent the algorithm from falling into local optimum, and improve the point cloud registration algorithm efficiency and precision.

Description

technical field [0001] The invention relates to machine vision and three-dimensional point cloud processing methods for workpiece size measurement, in particular to a point cloud registration method based on non-local operations. Background technique [0002] Point cloud registration is a very critical task in the fields of reverse engineering, unmanned driving, robotics, etc. It is mainly used to evaluate the geometric transformation between point clouds with unknown correspondences. The traditional point cloud registration algorithm is represented by Iterative Closest Point (ICP), which needs to iteratively find the corresponding point and calculate the minimum mean square rigid body transformation error, which limits the solution efficiency of the algorithm and is sensitive to the initial point cloud correspondence. , the algorithm is prone to fall into a local optimum. In recent years, deep learning methods have also been widely used in the field of point cloud registra...

AI Technical Summary

Problems solved by technology

Although the point cloud registration method based on deep learning gets rid of the dilemma of solving the geometric correspondence, its algorithm is limited to the feature extraction of the local information of the point cloud, and cannot fully consider the impact of the non-local point cloud structure information on the point cloud registration results. Influence, not conducive to the improvement of point cloud registration accuracy

Moreover, most deep learning-based methods still optimize the matching results by iteratively updating the transformation matrix, which increases the complexity of the algorithm and reduces the solution efficiency of the algorithm.

Images