A passive approach toward interactive 3D scene modeling

It is a great challenge to reconstruct a 3D real scene from image to graphics, using passive visual cues only. This paper describes a fully automatic interactive 3D model reconstruction system. The continuous video sequences help us to build precise inter-frame correspondences, and thus provide robust pose estimations. By fusing the silhouettes and the stereo components, a virtual object volume is obtained. A highly realistic 3D model is then synthesized by mapping the surface texture, and arbitrary views can be rendered via computer graphics techniques.