SPATIALLY DYNAMIC CALIBRATION OF AN EYE-TRACKING SYSTEM

ERICA is a personal-computer workstation developed to provide nonverbal, motor-disabled individuals with a means of communication and environmental control. The system employs computer vision to detect the user's approximate point of regard on a display screen and then invokes the menu option currently appearing at the corresponding screen coordinates. In this way, a severely handicapped user can run menu-driven applications with eye fixations as the only mode of input. This correspondence describes the design and testing of a spatially dynamic calibration for ERICA. The purpose of the new design is to improve the accuracy and precision of point-of-regard estimation by reducing the sensitivity of the system to head movement in the lateral plane (parallel to the display screen). The design introduces a second reference light source, which permits differentiation of head movement from eye rotation in the camera image. Symmetries in point-of-regard estimates for different fixed-head locations, which are observed by simulating the image response to eye and head motions, are exploited in the calibration model. The design is shown to provide accurate point-of-regard estimates in the presence of significant lateral head displacements while simultaneously improving overall estimation precision by approximately 85%.