Pedestrian Detection And Tracking Based On On-board Vision Systems

The dramatic development of road traffic provides a great convenience for people’sliving, but in the mean time, traffic safety issue has become increasingly prominent. Thedetection and tracking of pedestrian who are easier injured in the traffic become moreimportant to protect pedestrian safety. Therefore, the pedestrian detection and trackingbased on on-board vision is of great research value and application prospects. However, thedetection and tracking of pedestrian is difficult because of the complex environment and thereal-time requirements. At the same time, the serious block in the real scenario also bring agreat challenge to the pedestrian detection and tracking.For these problems, this thesis improves corresponding algorithms based on videocaptured by single-board camera, and thus forms a real-time and accurate pedestriandetection and tracking system. Major works are as follows:First, for the pedestrian detection system, acceleration methods of each link ofpedestrian detection in the framework of sliding window are discussed. Through the studyof the automotive vision, a new search method considering the pedestrian geometricconstraint is presented in this paper. This approach is based on the camera calibration todetermine the transformation relationship between image coordinates and the worldcoordinates. And furture, we can get the relationship between the pedestrian height and thepedestrian location. By this geometric constraint, the search band is dramatically reduced;Experimental results show that the proposed system can detect pedestrian based onon-board camera fast and effectively.Second, to make tracking more stable, the tracking by detection method is used in thisthesis, which can make the tracking robust by using the detection information. We realizethe pedestrian detection and tracking prototype system, and the system was tested on theCaltech pedestrian database. The results show that the constructed system have reached theapplication requirement of the automotive environment.