부산대학교 도서관

This work proposes an innovative approach for the study of the human face and its related gestures and movements, aimed to quantify the analysis of the health professionals. The developed system follows the movements of the face, and then relevant parameters and information related to facial gestures are extracted, which are useful for different healthcare areas (Psychology, Speech Language Pathologist). Most popular face models are designed to represent and simulate symmetric gestures. In this paper, we propose a set of modifications to the 3D Candide-3 model (which consists of asymmetric action units), in order to simulate gestures with asymmetry. Also, a combination technique between robust state-of-the-art facial landmarks and the modified 3D face model is employed for the facial points tracking phase, in order to improve the tracking performance. Encouraging results were obtained in the perceptual experiments performed.