Research On Moving Vehicle Detection And Tracking Algorithms In Video Surveillance

With the rapid development of computer technology,the intelligent transportation based on the image processing has been attracting more concern because of its feature of real time,accuracy and efficiency. Moving target detection and tracking algorithm can accomplish the process of traffic vehicles detection, tracking, classification and identification in the intelligent transportation. This thesis focuses on the algorithms of moving vehicle detection and tracking in the intelligent transportation system.To facilitate the post processing and improve the target recognition and tracking effect,this thesis studies some common image pre-processing methods, including the image restoration, grayscale, binary and morphological processing.On the basis of the analysis of the moving vehicle detection algorithms,the thesis simulates the optical flow algorithm on the basis of Lucas & Kanade model,the foreground detection algorithm on the basis of Code Book algorithm as well as the method of background subtraction on the basis of Gaussian.As the vehicle shadow affects the detection effect,this thesis studies the characteristics of vehicle shadow and eliminates the vehicle shadow.Through the analysis of the advantages and disadvantages of the above detection algorithm,the thesis proposes a new method of moving vehicle detection algorithm which combines the edge three-frame difference with the Gaussian mixture background subtraction. The simulation shows this method contains the full region and the complete information of the moving vehicle, which improves the effifiency and robustness.After researching and classifying the moving vehicle tracking algorithms, the thesis simulates the vehicle tracking based on the Kalman Filter and Meanshift algorithms;After analyzing the characteristics of the transport vehicles,this thesis establishes the feature space by the probability distribution of hue component for the simulation of the Camshift algorithm.Combined with the common problems of vehicle tracking in actual conditions,such as the vehicles’ occlusion,the similatity between the background environment and the objections, the similarity between the adjacent vehicles, the thesis simulates the method of combining the Kalman Filter and the Meanshift algorithm.