Fast lane detection & tracking based on Hough transform with reduced memory requirement

In this paper, we present a computationally efficient and robust lane detection algorithm based on Hough transform. The proposed method first extracts lane markings by applying 1D ridge detector to each row of an image. Given the extracted ridge points, the lane is then detected by applying Hough transform. Unlike the conventional methods, we consider only small set of line candidates which pass through a circle centered at previously detected vanishing point. The set is quite small and it is represented by a small region on a parametric domain. Hence, the proposed method needs much smaller accumulation array than conventional one and reduces the required memory. In addition, we propose modified parametric domain which encodes approximated lateral positions and the current lateral positions are searched on the modified parametric domain within certain region centered at previously detected positions. This proposed tracking scheme on the parametric domain considers temporal coherency and enables robust detection even when actual lane is weakly represented on the parametric domain. Experimental results show that the proposed method robustly detects the lane with reduced memory requirement.