Parsing main structures of indoor scenes from single RGB-D image

Abstract

Parsing indoor scenes from RGB-D imagery is a significant and challenging task in image understanding and computer vision. Since the walls, floor and ceiling are the main structures of indoor scenes, we propose a simple and efficient method to find them from a single depth image. We use the prior knowledge of the walls, floor and ceiling and distance histogram to obtain their planes, and delimit their extents by ray-casting. Our method can also detect the open windows or doors and other planes which have large area in indoor scenes. Experimental results show that our method is more robust to find the main structures of the indoor scenes than the existing method.