Occlusion detection and surface completion in 3D reconstruction of man-made environments

Abstract

This paper presents the methods for the quality improvement of the man-made environments reconstruction. The 3D model is created from the set of images without any user interaction. Some generic assumptions about the scene geometry are used; for example continuity of surfaces, piecewise planar scene or frequent parallelism and orthogonality between planes and lines. This information is employed to reduce the ambiguity of the 3D reconstruction and for the occlusion recovery. Method for the completion of the occluded surfaces is presented. First, the position of missing parts of the scene is detected and then the gaps are filled by the most suitable shape.