Registration of sets of points obtained by x-ray tomography with respect to CAD models

Abstract

The aim of this master thesis is to determine the defaults of an object by comparing a set of points obtained by its x-ray tomography (source) and a set of points obtained from its CAD model (target). The starting algorithm is made of three principal steps. First, the sets of points are pre-processed using a CAD software. Then, two registration steps are applied to the sets of points: namely, the covariance descriptor-based (CDB) algorithm and the improved iterative closest points (ICP) algorithm, with a novel estimation method for registration error proposed by G. Yao, Y. Zou, J. Wang, H. Yu and T. Chen, "Fully automated registration of 3D CT data to CAD model for surface deviation measurement". Finally, the error of each point of the "source" point cloud with respect to the "target" point cloud is displayed. Different improvements of the starting algorithm will then be tested in order to improve its performances and its robustness. The different algorithms are applied to artificial and real sets of points in order to determine which improvements to keep for the final algorithm. The criteria are a trade off between computation cost and robustness of the algorithm. The final result is a robust algorithm that automatically registers two corresponding sets of points, removes the non corresponding points between the two sets and finds deformations of the source point cloud. Other improvements that have not been tested in this study are then discussed in order to give ideas for the future improvement of the algorithm.