Multi-body ICP: Motion Segmentation of Rigid Objects on Dense Point Clouds

Abstract

Motion is one of the crucial keys for segmenting moving objects in dynamic environment. This paper proposes an approach of segmenting moving parts in the sequence of point clouds. Different rigid motions in dense point clouds are found and clustered by applying segmentation schemes such as graph-cuts into Iterative Closest Point (ICP) algorithm with initial segmentation from Generalized PCA (GPCA). Result shows that we can estimate motions of rigidly moving objects and segment parts having different motions simultaneously.