Registration process of geometric scan data of an object

Abstract

One of the main problems of laser scanned data processing is the problem of combining object's point clouds received from different angles into a single coordinate system. This process is called registration of geometric surfaces or relative aligning. The purpose of pairwise registration is to find a common contact surface and optimally align one of the scans relative to another. There are many algorithms for registration of two geometric surfaces that are the alternatives of the well-known Iterative Closest Point (ICP) algorithm. This article is concerned with the problems of using these algorithms and shows possible ways to solve them by evaluating the angular orientation matrix and the translation vector of the alignment transformation of the three-dimensional models. The article contains a description of a modified algorithm for the initial 3D models align approximation. The modified algorithm is intended to work with triangulated three-dimensional laser scans. Several insights of orientated gradients histogram building were provided for overlapping shapes. The resulted histograms were analyzed in order to evaluate the angular orientation, linear translation and rotation of the models.