Real-Time Multi-View Face Detection and Pose Estimation in Video Stream

Technologies for real-time multi-view face detection from video streams are indispensable to video content-based retrieval systems and video surveillance systems. In this paper, we proposed a solution for real-time multi-view face detection and pose estimation in video stream. Integrating both asymmetric and symmetric rectangle features, AdaBoost learning algorithm and pyramid like architecture is employed. Asymmetric rectangle features (ARFs) are inherited from symmetric rectangle features (SRF) to reasonably interpret asymmetric gray distribution in profile face image. Pose estimation for multi-view faces are brought out by view-based weighting algorithm (VB WA). Our primary experiments demonstrated that the system achieved high accuracy and high speed to detect both front and profile faces with their pose information from soccer video streams