Catadioptric line features detection using Hough transform

A line in space is projected to a conic in the central catadioptric image, and such a conic is called a line image. This work proposes a novel approach for efficiently detecting line images using Hough transform. Detecting line images brings two novel challenges for conic detection: one is that effects of occlusion are very significant where traditional conic detecting methods may fail, the other is that line images can belong to any type of conic, such as, line, circle, ellipse, hyperbola, parabola etc., and it is very difficult to detect a conic when its type is unknown. The main contribution of this work is that we prove a line image can be parameterized by only two parameters on the Gaussian sphere rather than five ones as a generic conic requires, and the above two challenges can be substantially solved accordingly. The validity of our proposed approach is illustrated by experiments.