Automatic 3D facial fitting for tracking in video sequence

This chapter presents the 3D face model initialization procedure. The major contribution is the innovative coarse-to-fine 3D initialization strategy proposed that deals with common problems of errors in feature selection process and with a lack of features points. The proposed approach is based over 3D deformable tracking system.We also propose a deep adaptation strategy while tracking for dealing with difference between real face mesh and the model ones. The 3D refinement proposed (also the deep one) are robust to initial light situation and expression.