A Head Pose and Facial Actions Tracking Method Based on Effecient Online Appearance Models

Target modeling and model fitting are the two important parts of the problem of object tracking. The former has to provide a good reference for the latter. Online appearance models (OAM) has been successfully used for facial features tracking on account of their strong ability to adapt to variations, however, it suffers from time-consuming model fitting. Inverse Compositional Image Alignment (ICIA) algorithm has been proved to be an efficient, robust and accurate fitting algorithm. In this work, we introduce an efficient online appearance models based on ICIA, and apply it to track head pose and facial actions in video. A 3d parameterized model, CANDIDE model, is used to model the face and facial expression, a weak perspective projection method is used to model the head pose, an adaptive appearance model is built on shape free texture, and then the efficient fitting algorithm is taken to track parameters of head pose and facial actions. Experiments demonstrate that the tracking algorithm is robust and efficient. Key–Words: Visual tracking, Online appearance models, Inverse Compositional Image Alignment, model learning, facial feature tracking