Study On The Vehicale Behavior Analysis Based On Video Tracking

In recent years, more and more traffic accidents are caused by the illegal behavior of vehicles. Therefore the behavioral analysis technology of detecting peccant vehicles has become a research focus. Most of the video-based detection technologies are based on object segmentation, while, the practical value of video-based tracking, which has more advantages in behavior detection, is reduced by the excessive time consumption.This paper mainly researches on the technology of vehicle behavior analysis based on video tracking for the purpose of practical application. Innovative research is made on basis of past researching.This paper improves the Moravec corner detection algorithm. The value of corner measurement is obtained by calculating the SAD between the areas around the current position. It can extract corners which are not sensitive to environment change. What's more, this algorithm is twice as efficient as the original algorithm. A dynamic algorithm is designed to select the optimal corners. It selects the optimal corners by changing the selecting window dynamically and can avoid selecting many corners on the same goal.According to the problem of large smooth information in the basic matching template, the new template consists of 6/25 of the unsmooth area in basic template. The template changes with the vehicles' position in tracking dynamically. This paper designs a fast matching algorithm. It uses Kalman filter to predict the predicted position and narrows the searching window to a quarter of the original size. Two different searching strategies are devised according to the smoothness of the existing track. The times of matching operation is reduced to a half of the total number of predicted position. The matching algorithm is many times more efficient than the FSA.A trajectory analysis algorithm is designed to analyze vehicle behavior. It detects the illegal retrograding by analyzing the angle between vehicle motion vector and the right driving direction. Illegal lane change is detected by analyzing the change of vehicles' position on road. It also detects abrupt acceleration and sudden braking by analyzing the change of vehicles' velocity change which is obtained by the change of vehicles' motion vector. The algorithm is tested on highway, urban roads, in tunnel, etcetera in most conditions, such as high noise environment, poor light environment, fog and snow environment, camera shading environment, etcetera. It is proved that the algorithm can work steadily in various conditions. The accuracy of event detection is more than 97%. The whole algorithm can meet the requirements of real-time detection.