Point cloud double-view-angle fine registration method based on projection from multiple constraint points to local curved surface

Abstract

The invention discloses a point cloud double-view-angle fine registration method based on projection from multiple constraint points to a local curved surface. The method includes steps of: calculating a multi-scale feature descriptor; determining an initial corresponding point set by using k-nearest neighbor search and overlapping rate parameters, and searching and screening corresponding point pairs by using feature descriptor similarity; performing curved surface projection on the corresponding area based on each point in the corresponding point pair to obtain a new corresponding point set; acquiring point pairs of bidirectional interpolation, carrying out screening by using a rigid transformation consistency principle, and solving a coordinate transformation matrix; solving of the coordinate transformation matrix and updating of the source point cloud are iterated in a loop mode till the convergence condition is met, and then the multi-view registration process is finished. The method is used for solving the problem of point cloud sparsity caused by discrete sampling of a measurement system which cannot be solved by an existing point cloud double-view-angle precise registration method.

Description

technical field [0001] The invention belongs to the field of reverse engineering, and in particular relates to a point cloud dual-view fine registration method based on projection from multi-constraint points to local curved surfaces. Background technique [0002] After the lidar scans the object from multiple angles to obtain a multi-view 3D point cloud, it needs to register the multi-frame point cloud data to obtain a complete object model. After using the coarse registration algorithm to roughly align the point clouds, it is necessary to further use the fine registration algorithm to improve the overall registration accuracy. Iterative closest point algorithm is widely used in fine registration. The basic principle is to continuously iteratively find the nearest neighbor points in the source point cloud relative to the target point cloud, determine the point-to-point correspondence between point clouds, and then calculate the covariance matrix of the corresponding point ...

AI Technical Summary

Problems solved by technology

[0003] The present invention provides a point cloud dual-view fine registration method based on multi-constraint point-to-local surface projection, which is used to solve the point cloud sparsity problem caused by the discrete sampling of the measurement system that cannot be solved by the existing point cloud dual-view fine registration method

Images