人臉偵測與辨識系統設計應用於錄影影像之搜尋

Abstract

現今的人臉偵測及辨識技術發展已相當多元化，從早期的靜態圖片分析到即時影像截取，並發展至目前3D深度影像的應用，舉例如臉書(Face Book)上的身份標籤及年齡辨識、手機人臉認證資訊防護、車牌辨識及門禁監視系統等，均基於此一技術上的延伸應用，本論文主要研究目的即設計一人臉偵測及辨識應用系統，搜尋錄影影像中人進入畫面時臉部正面出現的時間點並辨識此人臉的身份，可大幅縮短傳統採人工方式找尋錄影畫面的搜尋時間，同時統計出影片內所有進入過畫面的總人數，未來可進一步應用於相關的防盜監視系統或統計資訊中。 經實驗測試，利用上述的設計系統，在室內明亮的環境中可正確測得人進入監視器畫面時的正面人臉偵測率達65.38%，影片檔內每張圖片平均搜尋時間約為0.02秒，並獲得身份為訓練資料庫者之人臉正確辨識率達66.67%。

From earlier still image analysis to instant image capture and to 3D depth images application, the technique of face detection and recognition system is well established nowadays. For example, identity and age recognition on Face Book, mobile face verification systems, vehicle plate recognition systems, and access surveillance systems are all based on the application of the aforementioned technique. Therefore, the research purpose is to design a face detection and recognition system for detecting the appearance time point of a person in a video and for identifying that person. Consequently, we could significantly shorten the required time of video image searching and instantly count the number of people shown in a video image. In the future, we could even apply the technique to video surveillance systems and statistical information. With the system we designed, the test results show that the front face detection rate is 65.38% under bright indoor environment, the average searching time of each video image is roughly 0.02 sec., and the identity belong to training database people can get front face recognition rate is 66.67%.