多重攝影機之半自動化人物追蹤與特寫系統

Abstract

電腦視覺的領域中，攝影機的監控為一項重要的技術應用，利用多台攝影機監控場景時，需隨時注意各攝影機的變化，針對特寫目標的監控，導播無法同時專注於多台攝影機的畫面，在操作上可能需要隨時移動與縮放攝影機的畫面，進而影響到監控的效果或產生忽略的部分。本論文提出建立一個操作介面讓導播選擇特寫目標，透過ViBe背景模型、三維定位、相關評分標準、自動化特寫、攝影機的移動與縮放，可以讓特寫目標以適合的大小在各個畫面上顯示，達到同時監控的正確性與方便性。

In the field of computer vision, surveillance is an important application. For the use of multiple cameras to monitor the scene , director need to pay attention to the changes of the camera in time. To pick on close-up views, director cannot simultaneously focus on more than one camera screen, but also may need to keep moving and zooming the camera screen, thereby affects the performance. This thesis proposes an user interface for selecting the target of close-up views. Then, through ViBe background model, three-dimensional positioning, the relevant assessment criteria, we developed automated functionalities for controlling multiple cameras for continuously tracking a target person and displaying the close-up view of that person.