Towards automated large vocabulary gesture search
Alexandra Stefan, Haijing Wang, Vassilis Athitsos · 2009
This paper describes work towards designing a computer vision system for helping users look up the meaning of a sign. Sign lookup is treated as a video database retrieval problem. A video database is utilized that contains one or more video examples for each sign, for a large number of signs (close to 1000 in our current experiments). The emphasis of this paper is on evaluating the tradeoffs between a non-automated approach, where the user manually specifies hand locations in the input video, and a fully automated approach, where hand locations are determined using a computer vision module, thus introducing inaccuracies into the sign retrieval process. We experimentally evaluate both approaches and present their respective advantages and disadvantages.