Abstract
Computer vision is the modality that most frequently comes to mind when talking about Assistive Technology (AT) for the visually impaired. If a camera “sees” what the eye can see, then a blind person should be able to access visual information, suitably digested by a computer program. Factual evidence, however, points to the contrary. Only a handful of AT systems based on computer vision have been marketed successfully, and the functions they perform are highly specialized (OCR, color tellers, bank note readers). In this talk I will ponder the pros and cons of computer vision as a sensing modality for AT. The discussion will focus on issues such as reliability, hardware needs, and user interface. Case studies (both failures and successes) will be presented. Based on this critique, I will attempt a functional characterization of computer vision as a sensor for AT, and will highlight a number of application areas with realistic opportunities for this technology.