Parallel Evolutionary Registration of Range Data

Most range data registration techniques are variants on the iterative closest point (ICP) algorithm, proposed by Y. Chen and G. Medioni (1991, Proceedings of the IEEE Conference on Robotics and Automation) and P. J. Besl and N. D. McKay (1992, IEEE Trans. Pattern Anal. Mach. Intell. 14, 239—256). That algorithm, though, is only one approach to optimizing a least-squares point correspondence sum proposed by K. S. Arun, T. Huang, and S. D. Blostein (1987, IEEE Trans. Pattern Anal. Mach. Intell.9, 698—700). In its basic form ICP has many problems, for example, its reliance on preregistration by hand close to the global minimum and its tendency to converge to suboptimal or incorrect solutions. This paper reports on an evolutionary registration algorithm which does not require initial prealignment and has a very broad basin ofconvergence. It searches many areas of a registration parameter space in parallel and has available to it a selection of evolutionary techniques to avoid local minima which plague both ICP and its variants.