Video-based Automatic Incident Detection (AID) systems are widely deployed in many cities for detecting traffic incidents to provide smoother, safer and congestion free traffic flow. However, the accuracy of an AID system operating in an outdoor environment suffers from high false alarm rates due to environmental factors. These factors include snow movement, static shadow and static glare on the roads. In this paper, a robust real-time algorithm is proposed to detect snow movement in video streams to improve the rate of detection. This is done by having the AID system reducing its sensitivity in the areas that have snow movements. The feasibility of the proposed algorithm has been evaluated using traffic videos captured from several cameras at the City of Calgary.

Video-based Automatic Incident Detection (AID) systems are widely deployed in many cities for detecting traffic incidents to provide smoother, safer and congestion free traffic flow. However, the accuracy of an AID system operating in an outdoor environment suffers from high false alarm rates due to environmental factors. These factors include snow movement, static shadow and static glare on the roads. In this paper, a robust real-time algorithm is proposed to detect snow movement in video streams to improve the rate of detection. This is done by having the AID system reducing its sensitivity in the areas that have snow movements. The feasibility of the proposed algorithm has been evaluated using traffic videos captured from several cameras at the City of Calgary.