Eye tracking application for low vision

Abstract

Motivated from the result of our field work study that low vision people are hardly noticing public signs on the streets and in the interior of buildings even under clear weather, in this report we continue our research on application of eye tracking technology for low vision aids. We start from a short characterization of low vision from viewing standpoint and show that low vision person can basically recognize target object by his residual sight on his mobile display if we send an enlarged clear vision of the target. Taking advantage of this possible enhancement of low vision we explore eye tracking technology for helping him with navigation during this walking time. We show that classical scanpath technique for localizing regions of interest (ROIs) is applicable with low vision as well. Then we proceed to examine possible enhancement of low vision by (1) segmenting out public signs from his ROI, and (2) sending its enhanced vision back to his mobile monitor. We also show a preliminary result of public sign recognition in the view by using a fast pattern matching technique called “boosting,” liking to a future system of vision navigator for guiding the gaze of low vision to a missing public sign and zooming into it. Optimization of classifier programs is discussed from decision tree standpoint separately.