Improved 3D Model Search for Facial Feature Location and Pose Estimation in 2D images

摘要: This paper tackles the problem of accurately matching a 3D deformable face model to sequences images in challenging real-world scenarios with large amounts head movement, occlusion, and difficult lighting conditions. A baseline system involves searching set view-dependent local patches locate image features, using these update shape parameters. We show here two modifications that lead improvements performance can be applied other similar systems. These are: explicitly for occluding boundaries, which prevents from rotating rather than changing shape; simple method weighting relative importance each located match fit. demonstrate on both standard test sets series in-car driver videos, showing more accurate fewer search failures.