Continuous Hidden Markov Model based dynamic Persian Sign Language recognition

This study proposes a vision based Persian Sign Language (PSL) recognition system. Continuous Hidden Markov Model (HMM) with Gaussian mixture state observation densities is used to classify 15 dynamic signs. The proposed feature extraction approach is based on the spline interpolation of the sign trajectories. The efficiency of the system was assessed with a large set of videos collected by the authors. The system achieved the recognition rate of 98%. For signer-dependent and signer-independent experiments, the accuracy rates of 95.3% and 78% were obtained, respectively.