Skip to Main Content
Pedestrians are the most vulnerable urban traffic participants. In order to better protect them in pre-crash scenarios, it is necessary to detect them. Unfortunately, pedestrian detection is very difficult in highly cluttered urban scenarios, using cameras mounted on moving vehicle. A novel approach to walking pedestrian detection, using dense stereo vision systems, is presented. Multiple features combined into a Bayesian framework are used to yield a high rate of pedestrian detection. The feature set includes simple features such as width, height, lateral and longitudinal speed. It also includes complex motion features, such as the variance of the motion field caused by the pedestrians' legs and arms moving during walking and the periodicity of the pedestrians' walking pattern.