Improvement for 3D Coordinate Reconstruction of Facial Feature Points in Video Sequences

In this paper, accuracy improvement for reconstruction of three-dimensional (3D) coordinates of facial feature points in video sequences is proposed. While it is theoretically possible to reconstruct a 3D model of a rigid object from four feature points over three pictures, facial feature points in video sequences vary according to the facial expressions captured and the accuracy of reconstruction is not sufficient using the above-mentioned method. To improve the accuracy of the reconstructed 3D coordinates, we removed images with facial expressions including an open mouth or closed eye from video sequences and increased the number of images used to determine the feature points. Using information from 10 or more images, the proposed method can reconstruct precise 3D facial feature point data.