Study On Face Tracking Based On Compressive Sensing

The research of algorithm in face tracking is a hot topic in the Computer Vision and Pattern Recognition. With the development of field in Computer Vision, more and more researchers have focused on the problem of face tracking, mainly because of its broad application prospects, such as video surveillance. Meanwhile, face tracking combining with eyeball tracking could be applied to Fatigue Detection. The goal of face tracking is to find accurate face location in any background of any images, which could provide accurate face data for further face analysis. In the process of tracking face of videos, face tracking could be hard problems due to the variation of background of images in the videos. Moreover, the high dimensionality of face features extracted leads to large amount of calculation, which makes it hard to find face location by detection or tracking algorithm individually in the sequence of successive images. Although, researchers have done some experiments about face tracking, developing a robust and accurate algorithm of face tracking is still challenging because of light variation?face occlusion?face expression changing and etc. The thesis has studied how to design a robust and accurate algorithm of face tracking. The content is as follows:As for face images in the face databases, we firstly train face model based on facial points. In terms of thought of particle filter, we have proposed a method of extracting face features around facial points, which makes features extracted robust. Meanwhile, Compressive Sensing could project high dimensionality feature into low dimensionality space, which reduce amount of calculation.The thesis proposed face tracking method that is based on detection(Tracking-by-Detection), which adopts the strategy that more accurate means bigger weight and combines the results of detection and tracking to locate face in the images. This method improves the accuracy of algorithm and strengthens its generalization.The algorithm of face tracking has been tested by engineering samples and compared with recent algorithms of face tracking, which has demonstrated that our algorithm could be more robust and accurate.