Research On Point Cloud Matching Method Based On Dual Quaternion Algorithm

With the rapid development of computer engineering and extensive application, three-dimensional geometrical model has become the major digital media following the sound, images and video. Point cloud model is primitive geometric model based on discrete sampling points, is the natural representation of three-dimensional geometric model. Point cloud registration is obtained by feature matching of objects in the scene features and object model, Study how to raise from take implicit geometric information and topological information of scattered point cloud and the establishment of binding. Point cloud registration is to find a three-dimensional rigid-body translation and rotation so that the different perspectives of a common part of the point cloud to correctly match and overlap. point cloud registration get the more widely used in the three-dimensional computer-assisted surgical navigation, virtual reality, medical imaging, robotics and automation fields.This article study for three-dimensional point cloud registration algorithm and its related issues, and main tasks are as follows:1. The motion parameters estimation problem of the corresponding point set of is a core problem in point cloud registration. This paper presents an accurate and effective motion parameter estimation algorithm, is the dual quaternion method. Algorithm minimizing a objective function related to rotation and translation errors, Provides a new method to compute motion parameters .Experimental results show algorithm performance is better the SVD and quaternion method.2. This paper discussion the method to get curvature feature points, proposed a improved ICP algorithm based on curvature feature points. Algorithm using curvature instead of the point cloud based on initial registration and find nearest point using k-d tree, calculation of coordinate transformation use dual quaternion. Experimental results show that the algorithm efficient is better.3. This paper proposed a point cloud registration method based on octree. Algorithm break up the point cloud first, so that can compute curvature of point. then establish the hyopthesis relationship of point according to the corresponding the curvature of them. Algorithm avoids the time consuming of find nearest point in iteration. Experimental results show effectiveness of the algorithm.