Traffic Flow Detection Algorithm Research And System Implementation Based On The Video

Intelligent Transportation System (ITS) in recent years achieved rapid development in China. It can collect and process traffic information quickly, then command and dispatch traffic. ITS could play the maximum performance of the transportation infrastructure and enhance the quality of traffic management services.The video surveillance system is one of the most widely used technologies in the intelligent traffic system. Moving target detection and tracking based on video are the important parts of the video surveillance system, they prepare for the vehicle behavior analysis.Moving vehicle detection, tracking and traffic flow detection based on video are the focuses of this thesis. How to implement these algorithms in embedded systems is also the focus of the thesis. The main work is as follows:1. Moving target detection. The object of target detection algorithm is the video captured by the camera on the road to shoot the illegal vehicles. This article analysed four commonly used target detection algorithm based on the actual traffic video image, then it proposed a method of combining background subtraction based on improved kernel density estimation with improved hybrid frame difference for moving object detection. The method used improved nonparametric kernel density estimation to build the background model, and then realized background extraction and update. In hybrid frame difference, the method used dynamic threshold to extract the moving vehicle in the current frame. The parameters in the algorithm used several tests to determine. Experimental results were given to demonstrate that the proposed algorithm can improve real-time and accuracy in contrast with background subtraction and frame difference method.2. Moving target tracking. The article juxtaposed and analysed kalman filtering algorithm and GM(1,1) gray prediction model by the analysis of four commonly used tracking algorithm. It also compared and analysed Harris, Moravec and SUSAN corner detection algorithm, and then proposed a method of combining Harris corner with GM(1,1) model for tracking vehicle. The proposed method reduced the running time of the algorithm in contrast with traditional Harris corner detection algorithm. It also could track the target of the frame more accurately and quickly in contrast with Kalman filtering algorithm.3. Traffic flow detection. The paper detected the traffic flow of multi-lane with virtual line. It set up two virtual lines on the road, and then made the mean of the traffic flow through the two lines as the traffic volume on each lane. It could achieve more accurate traffic flow through the proposed method.4. The implementation of system. It used the embedded Linux system based on ARK-1310to transfer the collected video from industrial computer to the server. It detected, tracked the vehicle and detected the traffic flow in the received video by OpenCV. Experimental results were given to demonstrate that the proposed algorithm could deal with the actual traffic video in the system.