A vehicle recognition method robust against vehicles' overlapping based on stereo vision
Masatoshi Kimachi, Ying Wu, Satoshi Ogata · 2003
For the purpose of traffic measurement, it is necessary to recognize various vehicles, including those vehicles that are partly overlapped. A new stereo-based image processing vehicle recognition method which is robust against vehicles' overlapping, is proposed. A stereo-based method is selected because of its robustness to environmental changes. The proposed method overcomes the problem by three key points: 1) feature extraction based on a size-changeable feature extraction window; 2) camera calibration using environmental information; and 3) model matching with various overlapped vehicles' models. The effectiveness of the proposed method has been confirmed by experiments on real images. An average recognition accuracy rate above 95% was obtained for images of frequently overlapped vehicles.