A non-intrusive translation system to transform American Sign Language to digital text forms the pivotal point of discussion in the following thesis. With so many techniques which are being introduced for the same purpose in the present technological arena, this study lays claim to that relatively less trodden path of developing an unobtrusive, user-friendly and straightforward solution. The phase 1 of the Sign2 Project dealt with a single camera approach to achieve the same end of creating a translation system and my present investigation endeavors to develop a solution to improve the accuracy of results employing the methodology pursued in the Phase1 of the project. The present study is restricted to spelling out the American Sign Language alphabet and hence the only area of concentration would be the hand of the subject. This is as opposed to considering the entire ASL vocabulary which involves a more complex range of physical movement and intricate gesticulation. This investigation involved 3 subjects signing the ASL alphabet repetitively which were later used as a reference to recognize the letters in the words signed by the same subjects. Though the subject matter of this study does not differ by much from the Phase 1, the employment of an additional camera as a means to achieve better accuracy in results has been employed. The reasoning behind this approach is to attempt a closer imitation of the human depth perception. The best and most convincing information about the three dimensional world is attained by binocular vision and this theory is exploited in the current approach. For the purpose of this study, a humble attempt to come closer to the concept of binocular vision is made and only one aspect, that of the binocular disparity, is attempted to be emulated. The inference drawn from this analysis has proven the improved precision with which the ‘fist’ letters were identified. Owing to the fewer number of subjects and technical snags, the comprehensive body of data has been deprived to an extent but this thesis promises to deliver a basic foundation on which to build the future study and lays the guidelines to achieve a more complete and successful translation system.

Library of Congress Subject Headings

Optical pattern recognition; Image processing--Digital techniques; American Sign Language--Writing--Data processing

Publication Date


Document Type


Department, Program, or Center

Electrical, Computer and Telecommunications Engineering Technology (CAST)


Koont, Warren

Advisor/Committee Member

Indelicato, Mark


Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TA1650 .S37 2008


RIT – Main Campus