Traffic Condition And Abnormal Traffic Incidents Detection Based On Target Motion Trajectories

In recent years, with the continuous improvement of the computer’s ability of dataprocessing, and the rapid development of image processing and pattern recognitiontechnology, the traffic information detection technology based on video has become aresearch focus in the intelligent transportation field and attracted widespread attention. Themethods of the traffic condition and the abnormal traffic incidents detection are designed inthis paper for the purpose of practical engineering application.Obtaining good target motion trajectories is the precondition for the traffic condition andabnormal traffic incidents detection. The paper adopts the algorithm based on feature points toget the target motion trajectories. The motion targets are segmented by frame difference, thecorner points are chosen as feature points and then matched and tracked to form the motiontrajectories. The motion trajectories contain rich information about vehicle motion, such asspeed and motion direction of the running vehicle.Firstly, on the basis of the tracking trajectories, the algorithms about the trafficcongestion condition and abnormal traffic incidents detection for the fixed camera aredesigned. The over-speed and illegal parking events are detected by the changes of theinstantaneous speed of the trajectories; The retrograding events are detected by comparing theright driving direction and the vehicle motion direction; The illegal lane change events aredetected by the angle between vehicle motion direction and the fixed lane; The road statisticalspeed is used as the conversion condition to detect the traffic condition. Secondly, thealgorithms about the abnormal traffic incidents detection for the pan-and-tilt camera aredesigned. A lot of target motion trajectories are used to form the normal traffic mode of theroad, which is treated as the standard to detect various kinds of the abnormal traffic incidents.The algorithms proposed in this paper are tested in many scenes including urban roads,highways and tunnels. Moreover, the algorithms are also tested in the complex weatherconditions like sunny, rainy, cloudy days and the night as well as in the condition of thecamera wobble. The test results show the algorithms can meet the requirements of real-timeand accurate detection in various conditions, and the system works well with a high detection rate, it is of great research significance.