本論文提出一套具有人臉追蹤與辨識功能的數位智慧型監視系統,系統如果偵測到可疑人士入侵時,能及時發出警鈴與手機簡訊並儲存該重要畫面,以便取代現有類比式與僅做數位儲存的數位式監視系統。其中系統可分兩個子系統:人臉偵測追蹤以及人臉辨識。人臉偵測與追蹤是利用膚色範圍找出可能的人臉區域,以達到人臉初步定位的目的,並控制P/T/Z (Pan-Tilt-Zoom)攝影機追蹤該物體。初步定位後,再利用橢圓遮罩定位出人臉輪廓,最後利用眼睛與嘴唇偵測來確認入侵的物件是否為人。在人臉辨識部分,我們以二維小波轉換(2-D Discrete Wavelet Transform)取出人臉影像的低頻部分,這部份可以克服傳統影像特徵擷取的缺點,有效降低影像維度,另外利用線性鑑別式分析(Linear Discriminant Analysis)建立具有鑑別性之人臉模型參數。最後,我們使用最小歐式距離的決策方式,來判定誰是最有可能的人。 實驗結果顯示,在人臉定位部分,若在單純的背景環境中正確率為98.4%,在人臉辨識部分,單張影像辨識率達到94%。最後在P4 2G的電腦下,每張人臉影像辨識所花費的時間平均只需要0.26秒。
In order to substitute for the existing analog type surveillance systems having digital storage function, this paper presents a smart digital surveillance system with face tracking and recognition capability. If an invader is detected by the system, the system can promptly send out the alarm and SMS (Short Message Service), and store relevant image frames. The system can be divided into two sub-systems, i.e. face detection and tracking and face recognition. The human face detection and tracking sub-system identifies the possible face region using skin color information, achieving the preliminary positioning of the face, and control the P/T/Z (Pan-Tilt-Zoom) camera to track the object. After the preliminary face positioning we locate the face outline using an ellipse mask, and finally detect the eyes and the lip to confirm if the object of interest is indeed the human face. For the face recognition, we apply a two dimensional wavelet transform for the dimensionality reduction of face image. This method is able to overcome the drawback in traditional extraction of face features. In addition, we use the LDA (Linear Discriminant Analysis) to transform the features into a new space that has better separability. Finally, we employ the minimum Euclidean distance to determine the most likely person. The experimental results in face positioning part show that the successful rate is 98.4% in a simple background environment. For the face recognition part, the recognition rate for a single image reaches 94%. Finally, the computation time of the entire face recognition system is 0.26 seconds on the average, using a P4 2G personal computer.