3D Point Cloud Registration Algorithm Based On Plane Extraction

In recent years, with the rapid development of modern science and technology, 3D point cloud has gradually become appreciated and widely used in reverse engineering, heritage conservation, mining,3D printing and game entertainment. However, as the limitation of object size and the horizon of coordinate measurement equipment, point cloud data is unable to be collected through once scanning. Then, registration technology for point cloud arises. It unifies point cloud data from different horizons into the same coordinate system through the rotation and translation relations of different perspectives, in order to collect the whole point cloud data. This thesis aims to improve the precision and efficiency of point cloud registration.The research status of point cloud registration at home and abroad is investigated, and the typical registration algorithms in primary and precise registration stages are analyzed. Meanwhile, a series of classic plane extraction methods are introduced and compared in this thesis, such as region growing algorithm, Hough transform algorithm, random sampling consistency method, etc.Based on the analysis of typical registration algorithms, a new high-efficient registration algorithm is proposed in this thesis. Firstly, the algorithm divides point cloud date into source point cloud and target point cloud in different perspectives, then combines region growing algorithm and Hough transform algorithm to achieve expected efficiency. Secondly, under the guide of principal component analysis algorithm, plane polygons with suspected matching relation both in source and target point cloud are put into the same 2D coordinate system to calculate their mutual area ratio. In doing so, the plane polygons without matching relations are kicked out. Thirdly, utilize the theory that similar polygons have a similar centroid, this thesis will find the corresponding point in similar polygons of target point cloud with a certain polygon of source point cloud, then the primary registration begins. Finally, iterative closest point algorithm is applied to precise registration of point cloud.According to the contrast experiment of point cloud registration, this thesis will use the rock mass point cloud as the test data and compare with the typical algorithms, to test and verify the precision and efficiency of the new algorithm.The research indicates that the new algorithm has a great effect on objects which have large and massive structure surfaces. As a result, this thesis may be a certain reference for registration of huge rock matrix point cloud.