Szukana fraza: [Abstract = Method and tools of recognizing isolated words and singlesentences ofthe Polish Sign Language have been elaborated. Acanonical stereo system that observed the signer from a ITontalview was used. 14 feature vectors taking into account informationabout the hand shape, orientation, and 3 D position with respect tothe face were examined. 10 I words and 35 sentences that can beused at the doctor's and at the post office were recognized. Featurevectors were computed on the basis ofhuman skin detection anddisparity maps. Hidden Markov ModeIs were used for recognition. A data set of 6060 signs and 1400 sentences signed by two personswas prepared. Comparative experiments with regular and parallelHMMs were carried out. Recognition accuracy of about 98% hasbeen obtained. This thesis constitutes a step to provide amechanism of transcription sign language into text or speech. Furthermore it can serve as a good basis for the development ofgesturaI human\-machine interfaces. Additional result is a vision\-based system that makes continuation ofthis research possible. Data sets prepared for experiments are accessible via Internet.]

Wyników: 1