Developing a Real-Time Identify-and-Locate System for the Blind

It is safe to say that computer vision holds great potential for providing a broad range of benefits to the blind in the foreseeable future. But despite the rapid advances in computer hardware and vision algorithms, robust, self-contained functional systems that can be used by the blind for ‘identify-and-locate’ tasks are not yet available. This paper describes the challenges encountered and experience gained in building such a functional ‘identify-and-locate’ system by a group of undergraduates in the Gemstone Program at the University of Maryland, College Park. The greatest resistance to getting this task off the starting block was the need to simultaneously tackle several technical challenges requiring a range of expertise. These included developing a robust and real-time computer vision algorithm, a voice recognition and speech output system, a directional feedback interface, and the means to integrate these components into a single functional unit. With access to a pool of students with differing skills and backgrounds, we overcame the initial resistance. The bulk of the project was completed over a one year period, resulting in a prototype system that the blind can use in controlled environments to reliably identify and locate objects and signs within seconds.