Face Tracking Using Optical Flow Real-Time Optical Flow Enhanced AdaBoost Cascade Face Tracker

This master thesis deals with real-time algorithms and techniques for face detection and face tracking in videos. A new approach is presented where optical flow information is incorporated into the Viola-Jones face detection algorithm, allowing the algorithm to update the expected position of detected faces in the next frame. This continuity between video frames is not exploited by the original algorithm from Viola and Jones, in which face detection is static as information from previous frames is not considered. In contrast to the Viola-Jones face detector and also to the Kanade-Lucas-Tomasi tracker, the proposed face tracker preserves information about near-positives. In general terms the developed algorithm builds a likelihood map from results of the ViolaJones algorithm, then computes the optical flow between two consecutive frames and finally interpolates the likelihood map in the next frame by the computed flow map. Faces get extracted from the likelihood map using image segmentation techniques. Compared to the Viola-Jones algorithm an increase in stability as well as an improvement of the detection rate is achieved. Firstly, the real-time face detection algorithm from Viola and Jones is discussed. Secondly the author presents methods which are suitable for tracking faces. The theoretical overview leads to the description of the proposed face tracking algorithm. Both principle and implementation are discussed in detail. The software is written in C++ using the Open Computer Vision Library as well as the Matlab MEX interface. The resulting face tracker was tested on the Boston Head Tracking Database for which ground truth information is available. The proposed face tracking algorithm outperforms the ViolaJones face detector in terms of average detection rate and temporal consistency.