3D body reconstruction using graph optimization

Three dimensional (3D) body reconstruction can be used in many applications. In this paper, we present a system that uses eight Kinect sensors to reconstruct the whole body. For each depth image captured by Kinect sensor, boundary points and the ones that are too near or too far away from the sensor are deleted. Then we propose an improved iterative closest point (ICP) based registration algorithm to align different point clouds. In order to minimize the accumulate errors in registration procedure, a graph optimization method is utilized in this paper. Experiments have been done to verify our algorithm. The results demonstrate that the proposed algorithm can produce accurate 3D models of human bodies.