Real time facial features detection using boosted classifiers and geometrical models

In this short communication, we present a four steps algorithms for fast eyes and mouth detection and tracking in videos. Attentionnal cascades of boosted classifiers respectively trained on faces, eyes and mouths are used as a detection step. Facial features are also detected using a Chinese transformation which results in a finer localization than the holistic approach of boosted classifiers alone. The complete system can process a video at about 20 frames per second (320times240) for a single 80times80 face in an indoor environment.