Multi-core Accelerated Acquisition Of Human Face And Facial Features From Video Sequence

Face detection technology in video surveillance play a key role in many areas, such as identity authentication, intelligent visual surveillance, video telephony, etc. While as an import part of face recognition, facial feature localization is attracting more and more attention.The main objective of this paper is to use human detection and face detection algorithms to automatically detect human faces in video sequences, and control the camera auto-focus to the face region, get a clear image of the human face, then the feature localization algorithm runs on it to extract the exact location of the facial features.Based on the above objectives, with human detection, face detection, facial feature localization as the core algorithms, and motion detection, motion tracking, motion estimation algorithms as preprocessing tools, and combines with the practical need, this paper propose a framework that can detect face and exactly locate facial features in surveillance video, the framework solve the problem that existing intelligent control system cannot get a clear face and facial features. Based on existing part-template matching human detection algorithm that used in static images, we combine it with motion detection and applied it in video detect, and this effectively improve the detection speed. The SRF (Scattered Rectangle Feature ) based AdaBoost face detection algorithm effectively solved the high false positive rate problem of basic AdaBoost algorithm. We propose a facial feature localization algorithm that based on Gaussian derivative filter; this algorithm can fast and accurately locate the facial features. We use multi-core related technologies on human detection and face detection algorithm to accelerate performance and improve the detection response time, reduce delays.Compared to other face detection intelligent monitoring systems, this system have the ability to detect long-distance humans and faces, and focus human face automatically to obtain clear images, and locate facial feature automatically on the face images.