Abnormalities Detection In Traffic Monitoring Scenes

Abnormalities detection is a critical research field of Intellegent Transportation System (ITS). It will serve as an important component of ITS when it becomes applicable. The Virtual Loop System is being widely used now, but abnormalities detection technique based on the analysis of moving objects'activities has not been available yet. The market demand for this kind of detection system is enlarging. This paper gives a detailed research of the Virtual Loop System for traffic flow monitoring, and the Irregular Trajectories Detection System as well.A simple Virtual Loop System has been set up to count the traffic flow and roughly evaluate the vehicle speed. The form of the Virtual Loops is three successive Virtual Lines for each vehicle channel. It's like an evaluation gate placed on the road. This system is computation efficient and does not involve complex reasoning. It can count the traffic flow accurately. This paper presents a virtual lines'states analysis algorithm based on time window to improve the system performance. It holds the compute efficiency of the system, and does not introduce too much complexity.This paper gives a nuanced discussion of the Irregular Trajectories Detection System, and completely presents the algorithms of the four components, including ISM (illumination significance measure) extraction, foreground processing, objects tracking and trajectories analysis. The first two components contain the ISM thresholding method for extraction of the foreground points. It keeps the computation efficiency and can immediately respond to the sudden light changes of the scene. So it can meet the demands of real-time monitoring. In order to rule out the false-positive foreground points and parts of the shadow points, an algorithm for adjusting the bounding-box of the objects has been proposed. It can get better bounding-box, usually more accurate. This paper chooses the Mean Shift algorithm for objects tracking. At last, the principles of trajectories analysis are presented, including the Points of Interest and the concept of Centroid-Envelop. This paper proposes a POI (Points of Interest) Automated Extraction Algorithm. The Irregular Trajectories Detection includes two parts:real-time evaluation of the objects'tracks; the classification and storage of a certain entire track. Real-time evaluation mainly deals with the lane assignment. The entire track will be classified as normality or annormality. Then the detection results are presented.