Spotting Analysis of Sequential Representation of Japanese Finger-spelled Sign Language with RGD-D Camera

Abstract

Finger-spelled sign language recognition is one of hand sign recognition tasks and one of most challenging ones in some research area such as computer vision and human computer interaction. In this research, we especially focus on JFSL (Japanese Finger-spelled Sign Language). To realize a practical system for automatic finger-spelled sign recognition, we need to analyze the sequentially represented signs in real-time with some complex processes such as the recognition of signs one by one, word recognition, morphological and sentence analyses, etc. Although many finger-spelled sign language recognition researches have already been existed, they mainly focus on the one-by-one sign recognition and discuss the analysis of sequentially represented signs insufficiently. Therefore, compared with them, we focus on the spotting system of sequentially represented signs to move the next step of one-by-one sign recognition. In this paper, we propose a new spotting techniques for JFSL with SVM and rule-based approaches, which are based on moving direction of hand and finger directions before and after hand movement. From the experimental results, we achieved the 81% of average division rate for division of sequentially represented signs into one-by-one signs.