Facial expression has an important role for natural interaction among social robots and humans. In this paper, an architecture conceived for imitation of facial expressions is proposed. We describe the computer vision algorithm that was implemented for real-time geometric facial features extraction. It covers face detection, extraction of eyes, eyebrows, nostrils and mouth characteristic points, as well as an estimation of the face tilt angle. Fast processing has been prioritized over precision. Although we cannot make a definitive statement before all of the modules are built, the experiments show promising results on different subjects and illumination circumstances.