Fast Lane Detection Using Direction Kernel Function

In vision navigation tasks, lane marks on roads are very important vision cues. Usually, these lane marks can be detected using Hough transform or vanishing point detection. However such methods always need great computing power, and they are difficult to be realized on board processors in real-time. In this paper, a vanishing point like reference point is set as a support vector. The points on the lane marks can be easily clustered into one group by a certain distance definition with the support vector. Experiments results show that even with roughly selected reference point, the points on the lane marks can be easily detected in real-time. Another by-product of this method is that the information of lane marks of proceeding frames can be easily utilized during current process.