​● 中文摘要：

　　隨著科技蓬勃的發展，經由影像做人臉辨識的相關技術逐漸興起，且現今社會治安防護的問題依然存在，因此人臉辨識技術慢慢在生活上有很廣泛地應用。故本論文主要基於EmguCV實作校園犯罪預防之人臉辨識系統，一個類似保全的校園安全人臉辨識系統，用於解決校園安全問題，而該校園安全人臉辨識系統搭載於校園攝影機上加已提升校園安全犯罪的防護，在校園攝影機啟動的情況下，首先，系統會從攝影機獲取影像，並且擷取和記錄下這個人臉特徵，藉由訓練好的分類器來進行識別，當以後這個人進行識別和驗證時，會提出人臉特徵資料和現場擷取到的人臉特徵資料進行比對，當兩者的相似程度到達一定的閥值，就認證或識別出這個人的身份。然而，前面所述的技術僅於公眾場合環境下的識別，故本論文所要實現的是架設攝影機於校園環境，與前者相比，校園外環境下會遇到以下二種問題，第一種會遇到在人臉偵測後會因為光源變化、移動或遮蔽人臉導致人臉幀的遺漏問題，第二種則是在人來人往之情況下，攝影畫面偵測後辨識多個身份為「陌生人」的人臉問題，那會使得如果沒有在資料庫的人臉會一直判定身份為「陌生人」，因此，本論文加上了兩種方法，第一種方法則是加入了人臉追蹤技術在偵測辨識之後，減少了在攝影機範圍內偵測辨識完畢後因光源變化、移動或遮蔽人臉時，偵測人臉幀捕捉遺漏的問題，第二種方法則是陌生人可疑率，它會自動訓練身份為「陌生人」的人臉，訓練完畢後進行「陌生人」身份分類，再運用統計分析來判斷此身份為「陌生人」的可疑率與該陌生人的嚴重度。最後，本論文辨識人臉的方法以PCA、FisherFaces、LBPH實作出，並且可以用主動式的方式切換所想要的方法。

● English Abstract:

     Follow the rapid growth of technology,related techniques for Face Recognition have slowly risen. And with the persisting social security problems nowadays,Face Recognition started to widely applies in people’s life. Hence,This paper discusses about implementation of EmguCV,a face recognition system,in campus for crime prevention.When incorporated into the campus camera,enhances the campus security by identifying.The system will capture and record the person facial features via the campus camera for the use of trained classifier.These classified individual data of facial features is then used for identity recognition by comparing with image captured on-site.When the degree of similarity between saved and on-site data reaches a certain threshold,the person will be verified. However,the above mentioned technique only works well in public environment.Compared with the former,campus environment will encounter the following two issues.Firstly,light,movement or face being covered could render omissions.Secondly,the come-and-go of people will result in multiple identities being detected as "Strangers" which such faces will continuously judged as "Strangers" if they are not in face database.So, this paper added two technologies.One is face tracking technology that minimize omission issue due to light,movement,or covered face.The second technique is strangers suspicious rate.It automatically trains "Strangers" and do classification thereafter.Then uses statistical analysis to determine the severity of suspicious rate of the "Strangers".Finally,this paper based on face identification methods PCA,FisherFaces,and LBPH, and allow switching to the desired method in an active way.

● 文章連結：

http://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi/ccd=87IUHh/record?r1=4&h1=3

● 資料來源：

臺灣博碩士論文知識加值系統