Robust Infants Face Tracking with Active Appearance Model : a Mixed State Condensation Approach

This work addresses the problem of tracking, in real video sequences, the global motion of an infant face as well as the local motion of its inner features. This is a challenging task in Computer Vision field, because of the variability of facial appearance within a video sequence, most notably due to changes in head pose, expressions, lighting or occlusions. Thus, much research has been devoted to the problem of face tracking, as a specially difficult case of non-rigid object tracking. This task requires, by definition, the use of a model which describes the expected structure of the face. The advantage of explaining image data in terms of model parameters is to provide a natural basis for further interpretation, and it can be exploited by Human-Machine Interaction applications.