Deformable Mesh Model for Complex Multi-Object 3D Motion Estimation from Multi-Viewpoint Video

We propose a new algorithm using deformable mesh model for complex 3D motion estimation of multiple objects from multi-viewpoint video. In this paper, we define "complex motion" as motion which includes global change of the object shape topology. In complex motion, a part of the object may touch the other parts. To manage this effect, we introduce (1) "repulsive force" into deformable mesh model for simple motion estimation which integrates texture and silhouette information into unified computation scheme, and (2) efficient collision detection algorithm for deformable mesh model. Our deformable mesh model with repulsive force keeps hidden, collided surfaces to be touched each other, and gives dense, non-rigid complex 3D motion of the object. Some experimental results show that our deformation model can estimate motions of multiple objects and the object\´s motion with time-varying global topology, and gives topologically-consistent mesh models which can be compressed efficiently by conventional inter- frame 3D data compression algorithms and be used for 3D motion analysis.