Research On Scattered Point Cloud Registration Technology Based On Linear Constraint

In three dimension(3D) visual simulation of visual modeling, the reconstruction of the target object needs more groups of measuring data for registration. With the increasing maturity of the 3D scanning terminal, it has become reality that we have to obtain accurate local modeling data coordinates, the research of scattered point cloud registration is gradually becoming the key technology of 3D reconstruction. Automatic registration algorithms has been limited because of poor stability, low efficiency, so there are limitations in the application, using manual auxiliary matching methods in the commercial to realize the 3D data matching currently.This thesis focuses on data registration problems of the real object in the process of3 D reconstruction, get a new matching method based on the 4-points algorithm and the improved S-ICP algorithm. The method combines two aspects of global registration and local registration which can precisely match arbitrary starting position of two groups point cloud, obtained the high precision, a fast convergence and a relatively robust automatic registration method.Global registration is used to solve the problem of the point cloud initial matching in different position, in order to provide good initial values for local matching. In this paper, using 4-points algorithm which presented to achieve the value of global registration, this method is based on RANSAC scheme, and by using invariant features of geometric segment under the rigid transformation between four points, to improve the speed of registration algorithm. Local registration need an estimation of two point clouds position, and then achieve more accurate matching. Firstly combined with the algorithm Hong uses in the industrial applications to carry on the improvement of theobjective function, and combined with the current S-ICP algorithm, it is based on orthogonal force consistency singular value decomposition(SVD) algorithm, which improved the precision, convergence speed, convergence domain of the traditional ICP algorithm effectively. Based on the S-ICP algorithm, we have improved calculate the initial parameters, optimized the S-ICP algorithm and expanded the convergence domain of the ICP algorithm to a certain extent, improves the robustness of the algorithm effectively.The experimental results show that the automatic registration algorithm be able to obtain accurate results of more than 30% overlapped area of the point registration in this paper. This algorithm has good robustness, it is not only under the condition of the ICP algorithm and also has higher registration precision, even under the limited condition of the ICP algorithm can achieve registration accuracy efficiently.