| Summary: | 碩士 === 國立勤益科技大學 === 研發科技與資訊管理研究所 === 102 === Today, modern technologies are rapid developed and video surveillance systems are widely installed for public safety. The purposes of digital surveillance systems are to display and record video frames for real time analysis. The video frames are stored in a database system and used for tracking、computing and analyzing. The features, extracted from video frames, provide much information to administrators for decision-making.
Since most surveillance systems do not support intelligent detection function, it is not safe if a security guard operates a security system for a long time. It is impossible to work whole day long to security guards, who may persist fatigue to injure the person safety. The video, recorded from traditional camera, only provides evidences to someone without warring security guard on time.
The proposed system, Intelligent Face Detection Model (IFDM), is a human face detection system based on Haar-like features for detecting an illegal entrant at a restricted area. IFDM consist of Haar-like features, Adaboost algorithm, HOG features and Support Vector Machine which has provided the ability for a video surveillance system to detect human face. In IFDM, Haar-like features are used for describing human face features and provided the information for Adaboost algorithm to train and detect human face. The proposed algorithm, Hybrid HOG, can accurately analyze HOG features with the face area candidates from Adaboost. Finally, Support Vector Machine is used to classify faces and to locate the unique face area in a video frame. The experimental results of the study show that the accuracy rate of face detection at a restricted area can reach 92%. Thus, this dissertation proves that our proposed methods can be efficiently applied to the real-time face detection.
|
|---|
