Abstract
Structure From Motion (SFM) technique is usually used for camera motion recovery and 3D shape estimation. But the major problem with SFM is the occlusion of feature points which lead to hallucinate them, and thus increase the computation time. In this paper, we propose a method for a 3D shape reconstruction from a video sequence based on registering multiple partial reconstructions (patches). The proposed method avoids relying on hallucination step, which means a reduction of computation time. A realistic 3D textured shape is provided using a texture mapping pipeline based on the recovered motion of the camera. Experimental results on both synthetic and real images show that the proposed method is more than 200 times (on average) faster than the classical SFM methods which need to hallucinate the occluded points.