The Study Of Techniques On Segment And Tracking Of The Moving Vehicle In Vision

Traffic surveillance based on video technology is widely studied in recent years because of its easy-implementation and easy-maintenance. The vehicle tracking algorithms, which are the most important algorithms in traffic surveillance system, are the basis and premise of traffic information statistics and traffic incident detection.In this paper, the vehicle tracking algorithms are researched and analyzed. Main contents are included:(1) It presents algorithms of background estimation and image binaryzation based on block. The algorithms estimate background and binarize image for each block through calculating the mean and variance of a block.(2) After labeling vehicle with connected component labeling algorithm, it proposes a noise removing algorithm based on labeling area. Then the vehicle targets are merged by weighted average of abscissa and ordinate. Finally, the whole vehicle targets are extracted.(3) For the occlusion problem in tracking, it improves the association graph for vehicle tracking which was raised by Surendra Gupte, Osama Masoud and proposes a new tracking algorithm based on association trees. After the weight of overlapped area of two consecutive frames is calculated, the tracking tree is created of each target. It deals with vehicle occlusion problem better through the merging and dividing of the tree. Finally, the tracks of vehicle are calculated out with the trees.The algorithms proposed above are experimented in Second Ring Road of Xi'an where the traffic is often heavy. The experimental results show that it can deal with vehicle occlusion problem better. After analyzing the tracks of vehicle, abnormal tracks will be alarmed in time, such as incident, slow cars, stop cars and backing cars.