Establishing Robust Feature Point Detection and Tracking Methods for Face Orientation

Abstract

Recently, there has been an increasing need of a face alignment or facial feature point tracking for applying face recognition, facial expression estimation, face attributes prediction, and clinical face observation etc. This study proposes a robust facial feature tracking method against camera shake and face orientation changes. The method applies a cascaded composed learning (CCL) based facial feature point tracking method by incorporating optical flow for improving tracking accuracy and robustness. Experiments are conducted to show that efficient tracking is achieved by performing CCL combined with initial shape estimation via the optical flow.