Generalized cloud partitioning iterative closest point: uma avaliação quantitativa do registro de nuvens de pontos 3D

Abstract

In 3D reconstruction applications, an important issue is the matching of point clouds from different perspectives of a particular object or scene. Traditionally, this problem is solved by using the Iterative Closest Point (ICP) algorithm. To improve the efficiency of this technique, a methodology for reducing data sets in sub-clouds on the three orthogonal axes was proposed. An automatic convergence criterion was also proposed based on a micro-misalignment measure. In this work, the proposed technique was compared with ten other techniques. The results were evaluated using the RMSE metric, an analysis of the equivalent axis-angle representation of rotation, and the computational cost. The tests were carried out under ideal conditions and in conditions that simulate adversities, such as the existence of noise, rotations on a generic axis and the difference in density between the data sets. The experiments were carried out on several different data sets, acquired by several sensors, and revealed that the authorial approach achieved a more accurate cloud match, in a shorter time than other state-of-the-art techniques.