Vision-based autonomous detection of lane and pedestrians

We present an efficient approach to lane and pedestrian detection by processing sequential images from a camera attached to a moving vehicle. The left and right lines of the current lane are detected by finding high intensity pixels along multiple horizontal scan lines and connecting the detected pixel points. Line positions are predicted by tracking in order to increase detection credibility while reducing processing time. Pedestrian detection is done using HOG features. Since HOG-based method is however computer intensive, an edge based adaptive method is proposed. Our approach worked well on real road scene images.