Outdoor scene registration from 3D laser range data with coarse binary cubes

Outdoor scene registration of two 3D range scans, as usually required by field robotics, is very different from object and indoor registration, since all scan directions and depths may contain relevant data. This is not considered by 3D matching methods devised for computer graphics and reverse engineering. In this paper, we propose a specialized registration method based on the concept of coarse binary cubes within a cuboid fitted to the point cloud of the first scan. An integer objective function that does not employ point distances is defined as the number of coincident cubes. Its value is obtained by fast intersection of two binary vectors and does not involve any 3D data structures. This strategy avoids that denser range areas outweigh farther laser points, which can be useful to achieve good orientation estimations. The objective function is suitable for any optimization method. In particular, the Simplex algorithm has been adapted to avoid local minima. Experimental results in natural and urban environments are presented, where the method achieves similar accuracy and computation time as ICP, but increases robustness against initial misalignments.