隨著數位影像處理技術與DSP處理器的發展,大家對智慧型視訊監控系統的功能要求也日益增多;加上醫學的昌明,人類的平均壽命普遍增加,因而療養機構的需求也日漸增加。因此,國內外的產學界皆已投入大量資源在智慧型視訊監控系統的研發,並且致力將此系統導入長期療養機構中。因此本論文擬研發出一套低成本、移動便利性高並適用於DSP開發板上之移動物體的即時追蹤系統,旨在確實能達到目標物追蹤的功能,使看護人員可以更加方便進行監看。當有目標物進入時,系統在第一時間能即時偵測出,並使用PTZ攝影機加以追蹤,最後將目標物的影像輸出,以提供看護人員觀看,來降低看護人員的負擔與經濟成本。 本論文不同於一般的監控系統在於,我們結合了環場與PTZ兩種功能截然不同的攝影機,且未使用到相當複雜的演算法,大多利用基本的影像處理方法來達到即時監控的目的,而且本系統最主要是在DSP發展板上開發,並非在PC上模擬。而在偵測追蹤的過程當中,我們將環場攝影機所擷取到的連續環場影像,做前後張影像相減來取得目標物之所在座標,之後將環場影像中目標物的座標轉換到PTZ攝影機的座標,並控制PTZ攝影機旋轉至目標物的所在地點,來擷取目標物的影像。 由實驗結果顯示,我們利用兩種攝影機不同特性的優缺點做互補,且搭配上簡單的偵測與追蹤演算法,可以有效並即時的追蹤到移動的目標物,顯示本系統在一般環境的監控中是相當可行的。
Due to the technology advancement of digital image processing (DIP) and digital signal processing (DSP), the functional requirements of digital surveillance systems are more and more demanding. Moreover, since the average human life is increasing, we need more and more long-term sanatoria. Thus, a great deal of effort and resources have been given to develop a smart surveillance system, and try to apply this kind of system to long-term sanatoria. This thesis intends to propose a real time tracking algorithm of moving objects with low cost and high mobility features and the algorithm is well suited for DSP implementation. The goal of the resulting system is to ensure the availability of target tracking capability such that medical personnels can take care of the patient more conveniently. Our system can detect a moving object immediately when a person enters the surveillance area. Then the person is tracked by a P/T/Z camera, and the images of the target are sent to medical personnels for examination in order to reduce the loading of medical personnels and the cost. Our system is special became it combines an omni-directional camera with the P/T/Z camera, and no complex algorithms but some basic DIP methods are used to achieve our goal. Our system is developed mainly on the DSP kit, not on a PC. During the processing of tracking, we obtain the coordinates of tracking objects by subtracting the adjacent images captured from the omni-directional camera. Then we transform the coordinates obtained from processing the images captured by the omni-directional camera to the P/T/Z camera, and guide the P/T/Z camera to point to the target and capture the image containing the target. The experimental results show that by combining two complementary cameras with a simple detection and tracking algorithm, our system can track the moving target effectively and immediately, showing that our system is a feasible system.