Adaptive background updating algorithm for traffic congestion detection based on Kalman filtering and inter-frame centroid distanc

As current video based vehicle detection algorithms can not detect the traffic congestion accurately, this paper presents a new adaptive background updating algorithm based on Kalman filtering and inter-frame centroid distance. Firstly, a Gauss mixture background model is set up to extract the moving vehicles. Then, with Kalman filtering method, the moving vehicles are tracked to identify their motion states. This method predicts the centroid position of the next frame vehicles. The Euclidean distance of the centroids of the adjacent frames vehicles are counted and the appropriate threshold is set up to realize the identification and the mark of stationary vehicles in the video. This improved background updating algorithms can better judge the traffic congestion, and it lays a foundation for improving the accuracy rate of the detection of traffic flow. The proposed algorithm has been tested for multiple traffic videos. The results show that the algorithm is of good real-time ability, environmental adaptability and accuracy.