Generating 3D Faces by Tracking and Pose Estimation in Video Streams
Open AccessThis thesis presents a system that can generate animated 3D faces in real time, based on the ideas of Active Shape Model (ASM) and POS with Iterations algorithm (POSIT). The 3D face is controlled by human frontal face tracked in a video stream captured by commodity RGB camera, such as web cams and cameras on mobile phones. The pose of the human face is also estimated by analyzing the input video to create realistic shading on the generated face.Tracking faces in video streams is the estimation of the positions of a set of facial feature points in each frame. As human face is deformable, the relative distance between feature points varies significantly between facial expressions. The approach to tracking facial feature points is inspired by Active Shape Model (ASM) proposed by Tim Cootes. However, in the context of face tracking, their method lacks robustness for detecting feature points in motion and is not efficient enough to generate accurate estimations on-the-fly, an alternative method to fit face shapes to replace the original design in ASM. A series of experiments are conducted in this research and a critical analysis is provided. This thesis also discusses the impact of training data on tracking performance. The performance of generic shape models trained from a publicly available database are compared with its counterpart of person-specific models trained from the author’s self-portrait photographs.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.