人臉偵測應用經常搭配著人臉追蹤、人臉辨識等應用，作為後續應用程式的前置處理，本論文提出了一個可改善人臉偵測系統低硬體相依性、低電源供耗、高程式維護性性的人臉偵測系統。首先透過影像積分處理演算法將人臉影像積分求出人臉影像特徵向量，接著使用快速學習演算法篩選出人臉影像特徵，接著透過級聯分類器演算法中各個弱分類器比對加總並判斷出區塊中是否包含有人臉區塊，最後將上述各個演算法透過MIAT方法論將各個演算法分割為獨立的子功能模組，並透過GRAFCET建模工具替各個子模組建立離散事件模型，最後使用軟體合成技術將各個子功能模組獨立程式化，達到高移植性、高程式碼構架性。依實驗結果顯示在硬體相依性、電源供耗、系統架構化等方面皆優於傳統複雜龐大的人臉偵測系統，並透過此研究所提出的人臉偵測系統未來可更輕易的移植進各式嵌入式平台並可更輕易的結合各式應用如人臉辨識、人臉追蹤等。;Face detection is usually used in pre-processing of signals in applications such as face tracking and face recognition. This research proposes an enhanced face detection system that is of low hardware dependency, low power consumption and easy program maintenance. First of all, the Integral Image computation is used to derive the facial image feature vectors after which the Adaboost algorithm is applied to screen the facial image features. Then the weak classifier of the cascade classifier algorithm calculates and determines the area that contains human faces. Finally, the above-mentioned algorithms are divided into independent sub modules using the MIAT Theory. The GRAFCET modeling tool then builds discrete event model for each sub module. Finally, each sub functional module is written as an independent program so as to form a structure that is highly transferable and programmable. According to the experiment results, the hardware dependency, power consumption and system structure of the new system is better than the traditional complex face detection system. Hence, the face detection system proposed in this research can be easily integrated into all embedded systems and used in various different applications such as face recognition, face tracking and etc.
