本論文的目的是建立以電腦視覺為基礎的多汽車追蹤系統,可對同一汽車進行跨攝影機追蹤,因此可節省大量的人力資源,並且可以縮小事件的搜尋範圍。所謂的事件有肇事逃逸追蹤、在沒有車牌辨識下的贓車追蹤或是一些犯罪車子追蹤等等。藉由跨攝影機的追蹤知道該車開往哪個路口,就可以通知警方去做進一步的處理。 我們使用背景相減法取出前景物體,並對前景物體進行陰影去除、型態學以及凸包最小化。然後才對前景物體用bounding box distance追蹤物體。我們所使用的追蹤方法是將樣本取出特徵後,和目標物取出特徵做相似度比對,去算比對成功的機率,如果大於某個機率大小就比對成功。此方法比較不受場景的限制,也不需要事先訓練模組。; The aim of this thesis is to construct a computer vision-based multi-vehicle tracking non-over lapping system, which can track the same car across cameras. Therefore, a lot of human resources can be saved, and can narrow the search scope of the event. Events include a hit and run, stolen vehicle tracking or some crime tracking. By cross-camera tracking, which intersection the car bound for can be known and to notify the police for the further processing. First, background subtraction is used to extract foreground objects, then the shadow removing, morphological operation and convex hull is applied to make the objects more completely. Finally, the bounding box of the objects would be tracked. The tracking method used in this thesis is similarity matching of features, when the probability is higher than a threshold, the tracking is success. The limitation of the proposed mating method is relatively free from the scene, and does not require training in advance.