A new methodology for determining point-of-gaze in head-mounted eye tracking systems.

The ability to determine point-of-gaze with respect to an observed scene provides significant insight into human cognitive processes, since shifts in gaze position are generally guided by shifts in attentional focus. Using a head-mounted eye tracking system, a new methodology based on four or more point correspondences in two views was developed to reconstruct the subject's point-of-gaze. For exact point correspondences, 95% of the reconstruction errors are less than 0.32 degrees when the homography algorithm with distortion compensation is used to determine gaze position. In a typical visual scanning experiment, 95% of the reconstruction errors are less than 0.90 degrees. Analysis of normalization techniques that reduce the sensitivity of the homography algorithm to input errors suggests that the point correspondences should be arranged in a radially symmetric distribution around the area to be scanned. The new methodology was used in a clinical study on visual selective attention and mood disorders; this study showed that depressed subjects spent significantly more time looking at images with dysphoric themes than normal control subjects.