Abstract

We propose a method to recover the shape of a 3D roomfrom a full-view indoor panorama. Our algorithm can au-tomatically infer a 3D shape from a collection of partially oriented superpixel facets and line segments. The core partof the algorithm is a constraint graph, which includes linesand superpixels as vertices, and encodes their geometricrelations as edges. A novel approach is proposed to per-form 3D reconstruction based on the constraint graph bysolving all the geometric constraints as constrained linear least-squares. The selected constraints used for reconstruc-tion are identified using an occlusion detection method witha Markov random field. Experiments show that our methodcan recover room shapes that can not be addressed by pre-vious approaches. Our method is also efficient, that is, theinference time for each panorama is less than 1 minute.