Fast pedestrian detection based on Adaboost and probability template matching

In this paper, we propose a real time pedestrian detection approach which consists of two levels: coarse detection and further validation. First, partial stages of cascaded Adaboost classifiers are adopted to detect the upper bodies and generate candidate regions with a high detection rate. In the second level, a probability template is proposed, based on which a template matching technique is used to further reject the negative candidates. All the parameters involved are learnt from the training samples automatically. Our experimental results verify that the proposed approach improves detection performance substantially, while maintaining a fast processing speed.