Research On Real Time Vehicle Detection And Tracking Technology

As the safe city project drawing heavily in our country, the security monitoring industry evolving every day. As the key project of the Save City, the urban traffic attracted extensive attention of the whole society. Using the Intelligent Transportation System(ITS) to solve the problem of urban traffic has become a major method today. So as the basis of intelligent transportation to analysis traffic conditions, the moving vehicle detection and tracking become a hot spot.In this thesis we discussed video image preprocessing, vehicle detection and tracking, speed estimate and vehicle classification. Based on the analysis of the existing vehicle detection algorithm, finally we choose the background difference method as the algorithm of vehicle detection. By improving the Gaussian mixture model(GMM), the performance of the algorithm is as the same but the time of the computing is reduced. For tracking the vehicle, this thesis compared the Kalman filtering algorithm, particle filter algorithm and the CAMShift algorithm. Based on CAMShift algorithm, we used the Kalman filtering algorithm to track the vehicle. Experiments show that the algorithm has a good real-time performance. Considering the impact of vehicle classification, we proposed a algorithm that combining the virtual coil and the vehicle tracking algorithm to detect the speed of the vehicle. The algorithm start tracking the car when it pass through the virtual coil and stop tracking when leaving the virtual coil. Then calculate the time that the car used, as to calculate the speed of the vehicle. Finally detects the length of the vesicle to classify of vehicles.This thesis includes the following several parts:1.After the image preprocessing, we need to detect moving vehicles from the processing results. This thesis analyzes the average filtering, median filtering, Gaussian distribution model algorithm and compares the experimental results. Finally proposes a algorithm that can reduce the computing time without changing the detection rate.2.In order to analyze the traffic conditions, tracking the vehicle is very necessary. In this thesis we first introduce the current vehicle tracking algorithm. And then, according to the experimental result, finally chose to use higher real-time CAMShift algorithm with Kalman filtering to track the vehicle.3.At last, to estimate the vehicle speed and classify the vehicle, we use the combination of the vehicle tracking algorithm and virtual coil algorithm to estimate the vehicle speed and vehicle classification.