Estimation of pedestrian walking direction for driver assistance system
光哲 趙 · Medical Entomology and Zoology · 2012
I Estimation of Pedestrian Walking Direction for Driver Assistance System Abstract Road traffic accidents are a serious problem around the world, where the cost of human life is impossible to evaluate, and cause massive and continuous government spending. Different solutions have been proposed to reduce the effects of accidents, one of which, Advanced Driver Assistance Systems, as their name suggest, assist the driver by providing vital information on the traffic environment or by acting under specific circumstances to safeguard the occupants of the vehicle, or to facilitate driving. In case that an accident cannot be prevented, collision mitigation devices that are incorporated into vehicle design enhancement can be deployed to reduce the impact of the collision on the pedestrian. In this thesis we present a practical approach to the problem. Pedestrian protection is a crucial component of driver assistance systems. Our aim is to develop a video-based driver assistance system for the detection of the potential dangerous situation, in order to warn the driver. We address the problems of detecting pedestrian in real-world scenes and estimating walking direction with a single camera from a moving vehicle. The challenge is of considerable complexity due to the varying appearance of people (e.g., clothes, size, pose, shape, etc.), and the unstructured moving environments that urban scenarios represent. In addition, the required performance is demanding both in terms of computational time and detection rates. Considering all the available cues for predicting the possibility of collision is very important. The ―direction‖ in which the pedestrian is facing is one of the most important cues to predict where the pedestrian may move in future. Therefore we first emphasize the core problem of pedestrian orientation estimation in real-world scenes. The method is designed to estimate major eight orientations of different appearances. By taking account head-part orientation into the estimation, accuracy of overall estimation is drastically improved. Consequently, we construct and propose a three-stage method: (i) pedestrian detection, (ii) orientation estimation for single-frame and (iii) walking direction estimation for multi-frame. The first two stages employ and extend computationally