The Effect of Size and Position Normalization on HMM-based JSL Recognition
This paper shows a technique of continuous Japanese Sign Language (JSL) recognition based on Hidden Markov models (HMMs). The system aims for an automatic signer independent recognition of JSL sentences. Stereo video camera is utilized for extracting geometric properties of the head and hands trajectory. Size and position normalization is employed in order to account for differences in body-size and proportions. The system was implemented and evaluated using the RWC JSL database consisting of 64 sentences and 4 signers. The recognition results are then compared to the conclusion that normalization indeed increases the recognition rate of the system.
- 第7回情報科学技術フォーラム講演論文集, Sep. 2008
第7回情報科学技術フォーラム講演論文集, Sep. 2008, 549-550, 2008