Robust multi-directional bicycle recognition on the rotation using the in-vehicle cameras

Many researches have worked for the development of automotive technologies, so as to realize a convenient and safety automotive society such as Advanced Safety Vehicle (ASV) and Traffic Monitoring System (TMS). As a result, traffic accidents have decreased as the years pass, and the number of the dead and injuries have been reduced. However, autonomous driving is not only required to detect bicycles around vehicles, but also expected to understand the behaviors of bicycles. In addition, since the proportion of bicycle accident shows a high percentage in the total traffic accident, it has been predicted an increase of the accident of the bicycle in the future. Therefore we propose a multi-directional bicycle recognition system using the disparity information obtained by the stereo camera. In the proposed system, we divide the moving direction of the bicycle into three directions. In addition, we propose a robust feature extraction method on rotation in order to cope with the bicycle between each direction. Finally, we show the experimental results and verify the effectiveness of the proposed method.