Rapid Signer Adaptation for Isolated Sign Language Recognition

Research in the field of sign language recognition has not yet addressed the problem of interpersonal variance in large vocabulary on the classification level. Current recognition systems are designed for signer-dependent operation. Applied to signer-independent tasks, they show poor performance even when increasing the number of training signers. Better results can be achieved with dedicated adaptation methods. This paper describes a vision-based recognition system that quickly adapts to unknown signers. A combination of Maximum Likelihood Linear Regression and Maximum A Posteriori estimation was implemented and modified to consider the specifics of sign languages, such as one-handed signs. An extensive evaluation was performed in supervised and unsupervised mode on a vocabulary of 153 isolated signs. The proposed adaptation approach significantly increases accuracy even with a small amount of adaptation data. Supervised adaptation with 80 adaptation sequences yields a recognition accuracy of 78.6%, which is a relative improvement of 41.6% compared to the signerindependent baseline.