Scale and pose invariant real-time face detection and tracking

This paper extends the face detection framework proposed by Viola and Jones to handle all upright face poses between full left and full right profiles and also describes its combination with Camshift based face tracking system. The whole 180° degree range was divided into 5 sub-ranges and their corresponding view based detectors are constructed separately. Face detection framework is a combination of Real Adaboost algorithm, Integral Image and cascading classifiers. Faces are trained for five different poses (left, left+45°, front, right+45° and right) and face detectors are obtained for all poses. For tracking part of the system, Camshift algorithm is used because of its speed and easy implementation on video frames. To avoid impact of image analysis’s computations on Real-time application, multithreading methods are implemented by means of two processes. The combination of all these techniques yields good results for the face detection and tracking system.