Issues in fast 3D reconstruction from video sequences

In this paper, we discuss methods for incorporating data acquired as 3D surface scans of human faces into applications such as 3D animation and biometric 3D facial recognition. In both applications the challenge is to accurately and consistently find predefined features such as the corners of the eyes and the tip of the nose. In the field of biometry, if 3D face recognition is to compete with 2D methods, these features must be found to an accuracy greater than 1: 1000. In multimedia, the greatest problem occurs with animated 3D faces, where very small inaccuracies are clearly seen in moving faces. Therefore any inconsistencies must be found and rectified. Our work starts by providing a high-speed, accurate 3D model, and then developing methods to recognise the required features.