Alignment and tracking of facial features with component-based active appearance models and optical flow

The objective of this paper is to investigate alignment and tracking of facial features with component based active appearance models and optical flow. Face tracking becomes essential issue in the area of human-computer interactions. With accurate tracking and analysis of facial features, computers or robots can response properly to user´s facial emotional movements and facial expressions. We apply Active Appearance Model (AAM) and other techniques in real-time on cameras. Good AAM alignment results depend on proper selections of initial positions. However, it does take a lot of time when it applies image pyramid to get accurate results. In this paper, we introduce a brand new method to apply AAM fitting and effectively solve above problems. In our fitting scheme, we apply partial AAM separately on eyes and mouth. Thus, we can make facial features alignment much more efficient, and it becomes able to perform real-time alignment and tracking to real-world video. To make partial AAM more stable, we determine initial positions of facial feature models by multi-level optical flow. By the use of our developed algorithm, it is relative easier to get accurate positions of facial features and extract detail information of faces for further application in real world environments.