跨攝影機的車輛追蹤

Abstract

本論文提出一個跨攝影機車輛追蹤的方法，尋找車輛在不同攝影機間移動並予以標示。在本論文中，每個攝影機皆先經過校正時戳並相互有一小部分畫面區域重疊，其次對於每個攝影機的輸入影像都先使用背景相減法做處理，然後把得到的前景圖填入OpenCV Library裡的cvBlob類別作運算，可得畫面各移動車輛位置。最後再根據各道路方向計算欲鎖定車輛之移動軌跡並標示，並輔以車牌辨識資訊和絕對時戳資訊來推算其是否進入下一個攝影機的監控範圍，即可繼續計算其移動軌跡。

A method of vehicle tracking across different cameras is proposed in this thesis. We search the vehicles across the cameras and track them. In this thesis, each camera is supposed to synchronize its timestamp with the same time server, and covers a part of overlap area with each other. Second, we input the image to the background subtraction algorithm. Finally, we use the license plate recognition and timestamp information to derive the vehicle trajectory across different road intersections.