We propose to build a device that enhances the mobility of visually impaired persons by finding and reading signs aloud without the need for infrastructure beyond ordinary signs. Using new computer vision techniques, it will detect and read text in images captured by a camera worn like a pendant around the user's neck. We will build two commercially viable, self-contained consumer versions, a $1,500 device using consumer computers and cameras, and a $750 proprietary device. In typical use, a wearer will select a mode (city street, supermarket) by pressing buttons and optionally speaking commands, and then either point the device at a scene, or scan the scene using auto-repeat image capture. The device will find and read signs, but only output audio for signs relevant to the mode. The Phase II work plan has three tracks: 1) Computer vision software development and testing, 2) Human interface design and development, and 3) development and testing of the two forms of the device. Continuing our collaboration from Phase I, we use Smith-Kettlewell's Rehabilitation Engineering Research Center's expertise for human factors. Bolton Engineering will design and build the proprietary device hardware.
