关键词: 车辆检测, 交通流量监测, 背景学习, 运动分割

Abstract: A vision-based method of vehicle counting was proposed. Firstly the relatively static proportion of the video was extracted and the moving objects were segmented using differencing image algorithm. Then vehicles were tracked in divided road lanes. Compared with the existing schemes, motion segmentation in the proposed method was done by dynamic local thresholds. Moreover, a process of lane separator recognition was executed, which saved the time of vehicle statistics. Experiment results show that the new method performs better in recognition rates and average processing periods.

Key words: vehicle detection, traffic flow surveillance, background extraction, motion segmentation