Abstract
The iterative closest point (ICP) algorithm is a famous registration method that minimizes norms of difference vectors between two point clouds. ICP algorithm has widely applications. Thus, many researchers studied this algorithm. Recently, we see laser-scanned point cloud data in various fields. We call a point cloud target one, another point cloud source one respectively. The ICP algorithm calculates a rotation matrix and a translation vector from the former to the latter. The ICP must decide corresponding points of subset of target points. If both point clouds are noisy, the subset choose wrong corresponding points. We use locally fitted quadratic surfaces in order to get better corresponding points, which is effective. Hough transform is an algorithm to extract a feature curve in image analysis, computer vision, digital image processing. Hough transform uses voting procedure in parameter space to find certain classes of shapes that approximate noisy instances. In order to reduce noises, we project points to locally fitted quadratic surfaces estimated from Hough transform before ICP algorithm. As a result, we can calculate the transformation from ICP algorithm for noise reduced source/target point clouds. Hough transform is a robust algorithm. Thus obtained difference vectors are stable. Furthermore, precision of registration is better than those with original noisy point clouds.
