Active Appearance Model-based Facial-point Tracker (AAM-FPT): The AAM-FPT can be used to track 40 characteristic facial points shown in the Figure below. AAM-FPT consists of a hierarchy of three Active Appearance Models (AAM) and estimates the movements of the Head, Face, Eyebrows, Lips, Eyelids and Irises in 3D. Each AAM combines robust stochastic and deterministic appearance modelling with an optimized Levenberg-Marquardt Algorithm (LMA). The tracker has been implemented as a hierarchy of AAMs to enable handling self-occlusions, rigid and non-rigid motion. The powerful feature of the AAM-FPT is that it does not require prior training since it learns the appearance models on-line, while modelling outliers and dealing with illumination changes, occlusions, scale variation, out-of-plane movements, and translucent textures such as eyeglasses and sunglasses. The AAM-FPT is initialised automatically when a near frontal head pose is detected, using Active Shape Models (ASM) face alignment proposed in (Milborrow & Nicolls, 2008). The target of the initialization is to estimate the Deformation and Animation modes (SMD and AMD), which fit a Candide wireframe to the face within the face region detected by the face-detection module. The SMD and AMD parameters have been learned off-line from tracked sequences extracted from various public databases made available for research on automatic facial expression recognition.