Road Junction Background Reconstruction Based on Median Estimation and Support Vector Machines

Reconstruction of road junction background is the key for real-time traffic flow detection using background subtraction. High vehicle density and their rapid changing speeds at the junction present difficulty to background reconstruction. In this paper, we propose a new method that combines median estimation and the SVM theory for background modeling. We introduce a mechanism for collecting sample frames and blocks for background estimation. Background reconstruction based on Median values is efficient and accurate when the traffic flow is not severely congested. The SVM based background reconstruction method, on the other hand, can overcome the shortcoming of the median estimation method due to the traffic congestion and therefore enhance the reliability of reconstruction. The proposed method can be applied in intelligent transportation systems based on video sequences