A novel algorithm for automatic registration of multiple 3D point clouds

In this paper a fully automatic algorithm for registration of multiple 3D point clouds is presented. It is assumed that the data sets form a sequence and each data set in the sequence has an overlapping area with the previous one. The algorithm allows precise registration of such a sequence for data sets arbitrarily oriented with respect to each other. Thus, the algorithm is capable of aligning multiple point clouds as they are supplied by a 3D scanner, eliminating the need for manual registration or scanner and object position tracing. Genetic algorithm is used for preliminary pairwise alignment of successive views and at the same time for the overlap estimation. The registration is then refined using a variant of the ICP algorithm. At the last step, the pairwise registration error is diffused by a multiview registration algorithm. The proposed algorithm was tested on real measured data and proved to be precise, reliable and efficient