| Budget Amount *help |
¥2,900,000 (Direct Cost: ¥2,900,000)
Fiscal Year 2004: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 2003: ¥1,200,000 (Direct Cost: ¥1,200,000)
Fiscal Year 2002: ¥1,000,000 (Direct Cost: ¥1,000,000)
|
|---|
| Research Abstract |
Studies for development of non-contact, remote video-based, high-precision eye-gaze detection systems in which users can move their head around 10 cm were conducted for information input into a computer using light of sight. In the principle of the final eye gaze detection system, the precise 3-D pupil position was detected by two stereo wide cameras which can grab whole image of users' face. Another narrow camera fixed on a pan-tilt unit was directed toward user's eye, and grabbed a magnified eye image. In the image, the center of the pupil image and the center of the corneal reflection of the light source were detected. Using their relative positions, the angle of the light of sight relative to the light axis of the narrow camera was determined. The intersection between the light of sight and the visual object was detected as the eye gaze point.
Until 2003 fiscal year, the infrared LEDs were arranged in a ring shape around the aperture of each of the stereo cameras. The two rings of t
… More
he LEDs were switched on synchronously with an odd/even signal of the video and, the pupil and dark pupil images were alternately obtained in each camera. Diffentiating these images eased pupil detection. The 3D positions of the pupil was calculated and the narrow camera was made to direct toward the pupil. On the narrow camera, the rings of LEDs having a different wavelength from that of the stereo wide cameras were attached to near the aperture and a little far from there, and the pupil was detected in the same fashion, and the carnal reflection was detected in the bright eye image.
In 2004 fiscal year, an eye gaze calibration method was proposed, which utilized an assumption that the optics of eye ball has a shape of axial symmetry around the visual axis of the eyeball. This could easily calibrate eye gaze point by just looking at one known point on a PC screen and the narrow camera. The experimental results indicated that the detected eye gaze points showed the systematic errors in some positions of the screen. However, the change of the distance between the screen and the face of the subject did not influence the eye gaze detection results. This result showed the efibctiveness of this eye gaze detection method. In addition, it was difficult to estimate the pupil center in the image obtained from the narrow camera when the oarneal reflection was located at the edge of the pupil. The improvement of this problem made the dispersion of the eye gaze point on the screen less than 1.5 cm. Less
|
