A Consideration on Accelerating the Low-Overlap 3D Point Cloud Registration with CSVP

3D point clouds are represented object shapes in 3D space as a collection of points. The data are captured by an optical device, which eliminats the difference of the point density by the spesify of the device and limits the range measured. When we use data acquired from multiple directions, we need to registrate between 3D point clouds.However, if the overlap area is small, it is necessary to identify the overlapping areas. When we registrate between the low overlap point clouds, we put the merkers knowing the shape before and estimate the relation of the point cloud position from the relation of the markers. When we measure the shape of object where we cannot put the markers, it is difficult to registrate the point clouds.Therefore, our research objective was to develop a method to registate the 3D point cloud regardless of each point density. We introduced a weighted constraint function during patch selection for registration using the genetic algorithm of CSVP(Convexhull SuperVoxel Patch). Experimental results confirm the high registration accuracy equivalent to those of previous studies, even in the registration of 3D point cloud datasets that failed in previous studies. In addition, CSVP that were registreted by SVP(SuperVoxel Patch) were processed half time.