協同魚眼攝影機做多攝影機人物定位廣域監控系統

Abstract

多攝影機的監控系統擁有多個不同角度的視野資訊，可以在人物遮蔽的情況或者擁擠的場景中確實定位人物。然而在先前的系統中，每一個監控區域必須搭配四支傳統攝影機來架設系統，隨著監控區域的增加，需要使用的攝影機數量也會隨之成長，導致費用相當可觀。在考慮安裝攝影機的時間成本、設備成本下，本論文中提出一種方法，協同擁有廣大視野的魚眼攝影機與原系統做結合。我們首先校正魚眼攝影機的光軸垂直於地表後，接著同樣以基於消失點的線段抽樣方法，替代系統中的其中一支攝影機來完成人物定位。最後，我們分析魚眼攝影機所拍攝的影像特性，做出相對應的調整。實驗結果顯示，加入魚眼攝影機後，可以成功定位出人物，並降低多攝影機監控系統在廣域情境下的設備成本。

Multi-camera surveillance system can capture more rich visual information from different angle of views, which may locate people reliably in crowded scenes or when occlusion happens. However, a previous vanishing point-based system need to uses four cameras to monitor every area of concern. As the detection area increases, the number of such sets of cameras will raise in the same time. With the consideration of lowering the cost of equipment and setup time, we propose a novel way of using a fisheye camera in place of a traditional camera a multi-camera system. First, the fisheye camera is oriented so that its optical axis is perpendicular to the ground plane, and its principal center becomes the vanishing point of vertical lines in the scene. Then, the mapping between the image plane and ground plane, which is needed in the algorithm of people localization, is established by rectifying the radial distortion along these lines. Experiment result shows that we can indeed locate people suc-cessfully with a multi-camera system using a fisheye camera.