Probability Based Voting for Vanishing Point Detection

Abstract

This paper describes a method to detect vanishing points based on voting with probability. We use a group of lines in stead of a single line to describe the line passes a set of feature points belonging to the same line segment and propose a probability distribution to describe the group of lines. We use it to estimate the probability of all the points on the image plane that can be considered on the detected line. We use the surface of a cube centered at the projection center as the space for voting. We accumulate the probability of each point on the image plane (including the points at infinity) calculated with all the detected lines and project it on to the cube surface. The vanishing points are then detected by finding the cells that are local maximum of the accumulated probabilities and have many lines passed them. Through the comparative experiments, we confirmed that our method could give accurate and stable results and our method is not sensitive to the resolution of the voting space.