Robust face model based approach to head pose estimation

摘要

Head pose estimation from camera images is a computational problem that may influence many sociological, cognitive, interaction and marketing researches. It is especially crucial in the process of visual gaze estimation which accuracy depends not only on eye region analysis, but head inferring as well. Presented method exploits a 3d head model for a user head pose estimation as it outperforms, in the context of performance, popular appearance based approaches and assures efficient face head pose analysis. The novelty of the presented approach lies in a default head model refinement according to the selected facial features localisation. The new method not only achieves very high precision (about 4°), but iteratively improves the reference head model. The results of the head pose inferring experiments were verified with professional Vicon motion tracking system and head model refinement accuracy was verified with high precision Artec structural light scanner.