Assessment of human head pose in human-computer interaction

Abstract

A system is described that estimates a user's head position from a video image of it. The system includes three base algorithms: segmentation, detection of markers, and assessment of motion direction. The direction of head motion is determined by the dynamics of changing geometrical correlations between facial markers in the picture sequence. It is shown that the change in the angle formed by straight lines connecting the corners of the eyes and the tip of the nose have change dynamics similar to that of "yawing." This system operates in real time (7 fps) and ensures high precision in assessing the direction of motions (p = 0.95).