A parallel algorithm for 3D point pattern matching

The authors present a parallel algorithm for determining the correspondence between two sets of three-dimensional (3D) object feature points, referred to as frame 1 and frame 2, respectively. The points in frame 1 and frame 2 are obtained by observing the same dynamic scene (with multiple rigid objects), at two different instants of time. The parallel algorithm presented is adaptive, i.e. it works with a variable number of processors, and it uses a relatively weaker (and cheaper) model of parallel computation, namely, single-instruction multiple-data (SIMD) with shared memory blocks. The algorithm segments the scene based on difference in motion parameters. The approach is robust in the sense that it does not require the number of points in frame 1 and frame 2 to be identical