Extraction, Based On The Surveillance Video Of The Face Detection And Real-time Tracking

Face detection and face tracking are important research themes in the topic of Pattern Recognition and Computer Vision. They have a wide range of applications in many fields such as automatic human face recognition, video conference, intelligent video surveillance, advanced human-computer interaction, medical diagnosis, etc. After years of development face detection and face tracking have made great progress.For the development and research of face detection and tracking technology, we carry out a deep research and application into the face classifier using AdaBoost(Adaptive Boosting) algorithm based on Harr-like features; track the faces by Camshift algorithm(Continuously Adaptive Mean Shift). The main research innovations and contributions are summarized as follows:1. For the process of training classifiers and detection of AdaBoost algorithm, we improve AdaBoost algorithm as follows: 1) We simplify the multi-scale detection; 2) We train five cascade classifiers for multi-posture faces and connect them as a new face classifier. Then we achieve multi-posture face detection using this new classifier. Our system is able to detect most profile faces with the training of cascade classifiers of profile faces. Compared to traditional AdaBoost algorithm, our algorithm enhance detection capabilities.2. We deeply analysis the process of Camshift algorithm and we improve Camshift algorithm as follows: 1) We achieve automatic initialization of faces using AdaBoost face detection algorithm; 2) We combine face detection and face tracking. The result of face detection is the basis of face tracking while we detect the result of face tracking to make sure whether the tracking is correct or not and locate the exact location of the faces. This exhibits excellent performance in terms of tracking speed and accuracy.3. As a result of the actual needs of extract image of different angles from image sequence of faces, this paper presents and achieve several methods of roughly calculate the rotation of the faces, which we can achieve effective extraction of the faces.