Real Time Large Vocabulary Continuous Sign Language Recognition Based on OP/Viterbi Algorithm

Up to now, continuous sign language recognition is mainly based on statistical methods, especially hidden Markov models (HMM) and Viterbi-beam searching. However, the recognition speed often gets unacceptable with an increased vocabulary, which could cause a long time delay that is not fit for the real time recognition system. To speed up the recognition process, we present a method using one-pass (OP) pre-searching before Viterbi recognition. The experiments are processed in the large vocabulary database. Results show that the average recognition speed of OP/Viterbi approach can get a notable raise comparing with the single frame´s without reducing too much recognition accuracy