The Research And Design On Traffic Flow Statistic Based On Video Sequences

Traffic flow statistic is one hot and advanced topic in computer vision processing research, and also one core issues in intelligent transportation system (ITS). Traffic flow statistic can play a key role in reducing the urban traffic congestion and improving the utilization of the road.The main research contents of the traffic flow statistic are vehicle detection, shadow elimination and multi-object tracking. The ultimate goal of this algorithm is to improve real-time, accuracy and applicability of the traffic flow statistic. The main content of this article include the follows:(1) Introduce the traffic flow statistics by traditional method, such as detectors using toroid, ultrasonic. Meanwhile, the research on traffic flow statistics based on video sequences has developed quickly. Since it has many advantages, for instance, wider-area detection and superior flexibility, many research have been done in this field.(2) In terms of the moving vehicles detection, combine with background subtraction algorithm and three frame difference methods; use the OSTU dynamic threshold method to get the binary image after segmentation. Analysis the mechanism of shadow generating before suppress the shadow of vehicle, and according to the characteristics of shadow in colors space to eliminate the shadow in the vehicles. The vehicle is often close highly when detecting the moving vehicle. In order to separate the vehicle correctly, we use the method that make x-axis projection and y-axis projection of the vehicles' binary images.(3) In terms of vehicle tracking, we take the height, width, centroid coordinates and speed of the vehicle as feature vector, use the Kalman filter to forecast the location, height, width of the vehicle, so it can reduce the scope of search and the computation, and improve the real-time of this algorithm.(4) Analysis the component units of the system. In order to count the moving vehicles detected, we can statistics the counting real-time with the increase of the vehicles moving into the detecting region according to the tracking vehicles' ID in the window.