Fast 3d structure from motion with missing points from registration of partial reconstructions

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