Monocular Video Based High-fidelity Facial Animation Generation

Facial animation has a wide range of application in industry.But traditional facial animation making method is time consuming,laborious and costly.So finding a simple,convenient,low-cost animation making method is of great significance.This thesis presents a monocular video based automatic high-fidelity facial animation generation algorithm which can be divided into three parts:face detection,facial landmarks detection and high-fidelity facial animation capture.We start the algorithm by detecting and tracking important facial landmarks such as the mouth corners and eyes corners.In order to detect and track these facial landmarks,we need to train a face detector and a facial landmarks detector using Haar-like feature based Adaboost algorithm and shape-indexed feature based two-level booted regression algorithm respectively before the algorithm processes.Then we use detected facial landmarks to reconstruct 3D facial shape and pose in large scale for each frame.Finally,we utilize per-pixel shading cues to rebuild fine-scale facial details such as wrinkles and add these details to 3D facial shapes.