The VPLF method for vanishing point computation

摘要

A basic problem in computer vision is the extraction of 3D vanishing points (VP) out of 2D perspective projections. VP analysis provides strong cues for inferring the 3D structure of a scene from only a single view. In this work we present a novel method for VP computation based on a likelihood function that characterizes the plausibility of a point as VP. The performance of our method has been tested with a variety of artificial and real scenes and has been found very promising.