A Stereo-vision System for the Visually Impaired John Zelek, Richard Audette, Jocelyn Balthazaar, Craig Dunk School of Engineering University of Guelph Guelph, ON, Canada, N1G 2W1 Abstract Navigation is one significant barrier to individuals with vi- sual impairments. In response to this barrier, we have in- vestigated a wearable stereo-vision system and a prototype system has been developed. Preliminary investigations have proven the viability of this device for navigation through simple hallway structures. The prototype is inexpensive and consists of two USB cameras, a “virtual touch” feedback system and a wearable computer (inexpensive laptop or em- bedded platform). The feedback system is used to relay vi- sual information via tactile feedback through the user’s fin- gers. Experimentation has shown the feasibility of this ap- proach and we soon plan on experimenting with the planned demographics for such a device. Depth measures were com- puted using a pixel-to-pixel correspondence method making use of an epipolar geometry constraint. Real-time operation was achieved by limiting the field of view to the effective field of view deemed necessary for navigation. Experimen- tation was conducted with a non-calibrated unit. The depth measures were crude, however calculations reveal that this inadequacy may be overcome by making allowances for epipolar uncertainty in the depth from stereo algorithm. 1. Introduction Sight and hearing, specifically sight, are considered the senses people make most use of in everyday life. It is a multi-disciplinary effort to develop devices for individuals who happened to be challenged in one of these senses. A beneficial aid for a visually challenged person is one that fa- cilitates mobility independence. This paper presents some preliminary work in the development of a wearable device that uses inexpensive USB cameras, a simple tactile glove and light-weight computing facility to provide obstacle in- formation to a visually challenged person. The interest with respect to stereo vision is that it is rare that final system needs, constraints and requirements are reflected in the orig- inal depth-from-stereo algorithms. Albeit, we are only now at the stage in the project where we can make appropri- ate adjustments to the depth-from-stereo algorithm we used. This is as a result of a better assessment of the requirements of the algorithm in the context of the entire system. We used a unrefined standard algorithm and upon constructing a pro- totype, we are now better equipped to re-iterate our design, in particular, the design of an appropriate depth-from-stereo algorithm. 2. Background The work in vision substitution (a brief review is presented in [1]) has focused on two main issues: (1) reading and writ- ing; and (2) obstacle detection and avoidance. The most well-known device for reading was the raised dot code developed by Louis Braille in the 19th century. Another invention, the Optacon, was a tactile 6 by 24 matrix of vibrating pins corresponding to the brightness patterns of a camera developed by Linvill and Bliss in the 1960s (this was discontinued as a product in 1996 by the company Tele- Sensory). There have been many initiatives to develop text to speech reading machines, especially by major computer companies such as Apple and IBM. The original reading machine was introduced in 1975 by Raymond Kurzweil. The role of obstacle avoidance is to present spatial in- formation of the immediate environment for improving ori- entation, localization and mobility. Electronic devices have been developed for this purpose and are typically known as electronic travel aids (ETA’s) or blind mobility aids. The two oldest aids for this purpose are the walking cane and guide dog. The walking cane is an effective mechanical device which requires certain skills of the person using it to interpret the acoustical reflections that continually result from tapping. The cane’s range is only a few feet (limited by the person’s reach extent and the length of the cane). Some people find the cane difficult to master or spend significant amounts of time in the learning phase [2]. The guide dog alleviates some of the limitations of the cane but little infor- mation regarding orientation and navigation is conveyed to the blind traveller. In addition, dogs require constant care and extensive training for both the dog and person [3]. Early devices relied on an acoustical sensor (i.e., sonar) or a laser light [1]. Unfortunately, power consumption is an 1