Head Orientation Estimation using Characteristic Points of Face

  • D. Dervinis Šiauliai University

Abstract

Model-based method for estimation 3D head position is proposed. We used monocular web camera (camera maximum resolution 640x480 dpi). First, we created head model, where 3D rotation center is located between eyes. Second, by method OpenCV characteristic points of face are detected and continuous real time tracking was realized by normalized correlation. Rotation of head is calculated using Euler rotation matrixes. Equation system is solved by minimization method. In result we got three angles ϕ (roll), Θ (yaw) and ψ (pitch), which characterized 3D head rotation The experimental results show sensitivity to image noise. If noise level with standard deviation σ=1 is added, the estimation errors are less than 1 degree, if σ=2 – less than 2 degrees (resolution 320x240 dpi). Ill. 6, bibl. 11 (in English; summaries in English, Russian and Lithuanian).

Published
2015-03-20
Section
T 115 MEDICINE TECHNOLOGY