Tracking a head and estimating its pose with facial feature points

摘要

This paper describes a method for tracking facial motion and estimating head pose with facial feature points. The color information (the skin color and the hair color) and the edge information (the strength and the orientation) are extracted from an input image, and are used for detecting the regions of face and facial organs. The facial feature points are detected from those regions with SUSAN corner detector. The facial motion is tracked by calculating the correlation of the edge information in the rectangle range centering around the feature point for two consecutive input images and detecting feature points. Using those feature points and assuming weak perspective projection, we can estimate head pose with 3D head models made by Cyber Ware Rapid 3D Color Digitizer.