FACE TRACKING USING COLOR, ELLIPTICAL SHAPE FEATURES AND A DETECTION CASCADE OF BOOSTED CLASSIFIERS IN PARTICLE FILTER

This paper addresses the problem of tracking a face in images coming from an active camera. An improved on-line tracking technique that uses face detection results is proposed. The tracked face is represented by a weighted histogram reflecting both the color and the shape of the target. The histogram representing the tracked face is compared to histograms obtained at the positions of samples. The dissimilarity of the compared histograms is measured using the Bhattacharyya distance. An ellipse with fixed orientation is utilized to model the head and to extract the color distribution of the ellipse's interior. The color histogram and parameters of the ellipse are dynamically updated over time to generate object hypotheses and then to verify them by extracting the face candidates at the positions of samples. The integration of color cue and edge strength along the elliptical head boundary is done within the particle filter in a probabilistic manner. The filter assigns smaller weights to samples that are farther away from the centers of detected faces. A fast and efficient method of detection which relies on the AdaBoost algorithm and a set of Haar Wavelet like features is used to locate faces during tracking and to modify weights of the particles accordingly.