A Feature Extraction Approach of Traffic Congestion from Video

To avoid the difficulty of collecting accurate traffic flow data, this paper proposes a novel approach for congestion features extraction from traffic video. The approach firstly segments the traffic video into shots and the shot motion content feature is extracted. Then, we extract the key frames applying an improved global k-means clustering algorithm. The last congestion feature of the global optical flow energy is computed based on the key frames. The numerical experiments on traffic surveillance video show the validity and high accuracy for traffic congestion detection using the propose method in this paper.