監視器系統隨著數位化的演變,許多技術被應用在這個領域。然而監視器也因應社會的變遷不斷的增設,而這樣的需求是無可必免的,主要是要嚇阻犯罪的發生和事後尋找犯罪的證據。 本論文主要目的是開發一個能夠掌握特定目標,在多監視器架設的區域內移動的系統。其中使用貝式模型來建構前景物件偵測的方法,將在監視畫面中在移動的物件抓取出來,並對感興趣的目標以Mean-Shift做追蹤。接著利用該目標出現的特定時間和地點,在下一個可能會出現的鏡頭找尋目標。找尋目標過程中,使用HSV色彩空間計算直方圖,再和目標比較其相似性。 研究中實驗的環境為戶外的監視器畫面。總共抓取三個鄰近的監視器。依照使用者選取的特定物件,對該物件的出現時間、地點和色彩特徵做追蹤,將此將追蹤的結果提供給監視人員。 本論文貢獻是提供能夠在多監視器系統追蹤特定目標的系統,並討論在實現本系統所遇到的問題, 以及解決的方法。
This thesis presents a system for tracking a target of interest across in multi-camera system. The analysis includes three parts: the first part is object segmentation by Bayesian model. The second part is object tracking. Using the object segmentation results and Mean-Shift to track the target we interested in current camera. Last part is collaborate information of each camera for tracking the target in multi-camera. Developing system provides users define the multi-camera system environment as they want, and video browsing interface lets users choose the interested target, finally showing the result helps them to know the target trajectories quickly. Experiment is used for three surveillance cameras in outdoor environment that recorded for one hour. And we will discuss the problems and solutions in realizing our system.