Fast automatic single-view 3D reconstruction of urban scenes

Olga Barinova
(Moscow State University, Russia, obarinova@graphics.cs.msu.ru)

joint work with Vadim Konushin, Anton Yakubenko, Hwasup Lim, KeeChang Lee, Anton Konushin

a We consider the problem of estimating 3D structure from a single still image of an outdoor urban scene. Our goal is to efficiently create 3D models which are visually pleasing. We assume that the environment is made of a number of vertical walls and a round plane like in Hoiem et al, but introduce additional constraints on resulting 3D model structure. We achieve computational efficiency by special preprocessing together with stagewise search of 3D model parameters, dividing the problem into two smaller sub-problems on graphs with simple structure.

The use of Conditional Random Field models for both problems allows to capture appearance, geometric and context cues. Orientation of vertical walls of 3D model is recovered from vanishing points. Compared to Hoiem et al [1], proposed framework has lower run-time complexity, while producing 3D models of urban scenes of comparable or better visual quality.