Scale registration based on descriptor analysis and B-spline matching

By adopting different reconstruction techniques, 3D sensors generate different point clouds of various scales for the same object. This paper presents a new algorithm to align point clouds of different scales. First, shape descriptors based on different neighboring sizes are calculated on each point. Secondly, for each neighboring size, principle components analysis is applied on all descriptors to generate the cumulative contribution rates, which are fitted by B-spline curves. Then, finding scale different between two 3D point clouds in 3D space, is reduced to finding scale difference between two sets of 2D B spline curves in 2D space. Finally, a numerical solution is presented to find the unknown scale, which is experimentally proved to have higher accuracy than existing methods.