隨著電腦速度提升,即時人臉辨識或人眼追蹤應用在個人電腦上門檻越來越低。考量視訊攝影機(Webcam)產品價位低廉亦具備方便性與可攜性,筆記型電腦也普遍都有此裝置,研究成品應用性高,所以採用平價的網路視訊攝影機及一般個人電腦為研究平台發展人眼追蹤的相關研究。 本論文設計一個倒傳遞類神經網路產生每幀影像的二值化門檻值,將視訊影像二值化處理,突顯影像中人臉特徵。再以快速連通元件標記法標示二值化影像中各個圖塊元件,並計算其不變矩,與樣本不變矩比對找尋人眼圖塊元件。最後在尋到的人眼影像上,繪出上下眼瞼的輪廓線,以兩眼角的角度作為判斷眼睛開闔程度的依據。經實驗結果證明,本系統可有效的進行即時的人眼追蹤以及眨眼偵測。
With computer process speed increasing, real-time face recognition, or eye tracking applications on personal computers getting lower and lower threshold. In this thesis, considering the video cameras (Webcam) are cheap, convenient and portability, thus the most of laptops have this device generally. So this study use personal computer with a webcam to develop the eye-tracking research. This thesis developed a backpropagation neuronal network to generate threshold value for each frame image. Thresholding video image could highlight face features in the image. Using fast connected component labeling to mark all components of a binary image, and calculating the moment invariants to search eye features. Finally, the upper and lower eyelid contours are drawn on eye images for judging the eyes are close or open. Experimental results show that this system can effectively carry out real-time eye tracking and blink detection.