PROPOSAL OF INFANT AUTISM SCREENING SYSTEM BASED ON GAZE POSITIONS IN FACE MOTION PICTURE
Kiyotaka Fukumoto, Yoshinobu Ebisawa
Shizuoka University, Japan
We propose a type of quick screening system for infant autism based on the recent finding that infants with autism do not tend to look into mother’s eyes, which performs by detecting in real-time where the infant tends to look at his/her mother’s face. 3D coordinates of the mother’s pupils are detected using a pupil detection system which we have developed. Two cameras in this system and a color camera are camera-calibrated. So, from the 3D pupil coordinates, the pupil positions on mother’s face motion picture being simultaneously shot by the color camera are estimated. The mother’s eyes- and the mouth-areas on the motion picture are determined based on the relative positions between the detected two pupils. The mother put the infant on her lap. The infants’ gaze positions on the motion pictures presented on a PC screen are detected using our gaze detection system. In order to compensate the difference between the optical and visual axes of the eyeball, a gaze calibration procedure is accomplished by looking at just one quickly shrinking circular target presented at the center of the screen before the screening. This gaze calibration method would function well even if it applies to infants because the target strongly attracts the infant’s attention onto it. The time ratios which the infant is gazing within the eyes- and mouth- areas are calculated automatically, and then are shown on the screen as autism indices. The time lag differences among the pupil and gaze detection systems and mother’s face presentation are compensated. The experimental results showed that the detected gaze positions of infant’s role healthy adult subjects concentrated in the small regions centering each of the eyes- and the mouth-areas of a mother’s role subject even when the latter subject was moving the head laterally.