Real-time Bicycle Recognition for Intelligent Rear Collision Warning Systems

摘要

Accurate and real-time recognition of pedestrians and cyclists have recently emerged as a premier research topic in the field of intelligent transport systems and computer vision. We present a vision-based framework for reliable recognition of cyclists in urban traffic environment with tight real-time constraints. The cyclist recognition completes our obstacle detection and recognition system composed of two fish-eye cameras invaluable for detecting side-appearing obstacles. The performance of this approach is evaluated and compared with recent publications using real-world data in various urban scenarios. Our experimental results suggest that it is possible to obtain high recognition rates (95%) and low false-alarm rates (less than 5%) by fusion of 3D and 2D data obtained with ultra wide-angle stereo images.