Registration algorithm research for work-pieces based on affine invariant

Abstract

Work-pieces registration is a crucial item in flexible manufacture assembly. In order to finish work-pieces registration, a method is proposed based on affine invariant. After the CAD model data and actual measurement data of a work-piece are acquired, the method consists of the following five steps: sampling point clouds data, extracting characteristic four-points set, searching characteristic congruent four-points set, computing transformation congruent matrix and the last, further precise point clouds data registration. Point clouds data of CAD and measuring model are sampled respectively through calculating the curvature of the two sets of data and selecting the obvious curvature points as their reduced characteristic points. Based on this, the characteristic four-points set from the reduced ones of CAD model and the corresponding matching congruent four-points set of the measuring model are extracted according to RANSAC algorithm. The rotation matrix R and translation vector T of any two matching four-points are then calculated through the algorithm of quaternion. After that, the measuring model is rotated and translated and then compared with the CAD model data, the most congruent transformation matrix is selected as the coarse registration result. Furthermore, Iterative Closest Point (ICP) algorithm is applied to the congruent transformation to improve registration precision. The experiment shows that the run time of the algorithm is 129.56s and mean-error of point-to-point distance is 0.062mm when accessing measuring model data more than 80000 points. Compared with the traditional curvature registration, the experiment also shows that the algorithm is more efficient and robust when the volume of point clouds data is larger.