Real-time active 3D shape reconstruction for 3D video

Abstract

In this paper, we focus on the real-time 3D shape reconstruction of an object moving in a wide area (e.g. a dancing human). To efficiently record dynamically changing 3D object shape, the real-time 3D shape reconstruction and active tracking are required. For the real-time 3D shape reconstruction, we propose a parallel plane-based volume intersection method. By implementing the algorithm on a PC cluster system, we have succeeded in real-time 3D shape reconstruction. Then to get the 3D shape of a moving object in a wide space, we introduce active object tracking by multiple cameras. To realize the active 3D shape reconstruction, we augmented the volume intersection method so that it can be applied to those images captured by active cameras.