Abstract:

Artificial scenes contain many parallel lines and orthogonal edges. These parallel lines projected onto lines in the image will meet in a common point. This point of intersection is the vanishing point. Vanishing point detection is of important significance for camera calibration, 3D scene reconstruction and so on. Traditional vanishing point detection algorithms are always based on 2D Hough parameter space. The complexity is high and the efficiency is low. Therefore a novel method is proposed. Firstly, long straight edges are detected followed by selecting and grouping these lines. Secondly, the constraints of relationship between vanishing point and focal length are used to get specific location of the vanishing points. The method decomposes a 2D Hough parameter space into two cascaded 1D Hough parameter spaces. Experiments show that this method has lower complexity and shorter running time. In outdoor artificial scene images, it performs more robust and accurate.      

Key words: vanishing point;vanishing line;camera calibration;line detection