Pedestrian Detection in Traffic Scene through Cascade AdaBoost Classifier and Target Tracking

摘要

Pedestrian detection is an important component in an intelligent traffic monitoring system. In this paper, a real-time two-layer pedestrian classification method is presented. In the approach, firstly, several simplified geometric features are used to train the first layer pedestrian detector based on cascade AdaBoost, especially the 3D height geometric feature. Then, an improved multiple feature matching models is adopted for the tracking process and the actual motion trajectories in a 3D space are built. After that, the motion trajectories are dealt with using the linear fitting method based on subsection process to form the final subsection trajectories. Finally, the second layer pedestrian detector is applied on the subsection trajectories for the further pedestrian detection. Experimental results show that the improved method in this paper has a high correct detection rate, as well as a fast detection speed.