A robust traffic state parameters extract approach based on video for traffic surveillance

Vision-based sensors for traffic surveillance have attracted more attention because of their area sensing ability and flexibility. Conventional methods used to extracted vehicles mainly including background subtraction and images differences. Possible questions brought by these methods are time costuming and lack of robustness. Different from previous research, a new method based on wavelet transform is proposed. One dimension data set is generated from ROI of one frame in video. Vehicle edge is acquired from characterization of signals from wavelet transform. Furthermore, linear characterization used to represent edge of vehicles. Then combined geometry characterization and a dynamic criterion using histogram-based method are proposed to eliminate all unwanted shadow based on the linear characterization of edge. Speed of vehicles is obtained based on detective lines and minimum boundary rectangle (MBR), avoiding using Kalman filter to extract vehicle speed, reducing the huge computation. Experimental results show that the proposed method is more robust and accurate than traditional methods.