Robust pupil detection by image difference with positional compensation

Remote and noncontact, video camera-based pupil detection techniques are useful for human interface and human monitoring devices. In general, it is difficult to detect pupils without illumination. However, if using near-infrared illuminations, pupil detection becomes dramatically robust. We formerly proposed a robust pupil detection technique using two light sources and the image difference method. In the pupil detection technique, the two light sources alternately produce the bright and dark pupil images every field, and then the successive bright and dark pupil images are differentiated. In the obtained difference image, the pupil image can be easily detected because the background image except for pupils is basically canceled out. The above-mentioned pupil detection technique has some advantages compared to pupil detection techniques from the bright or dark pupil image; robust for high and dark ambient lighting conditions and for eye-glasses reflection, and so on. However, there is the following weak point: due to the time discrepancy in acquiring the bright and dark pupil images, its effect is spoiled when the pupils move. In this paper, three methods which improve this problem are proposed. The experimental results showed their effectiveness.