A Comparison of SimpSVM and RVM for Sign Language Recognition

Sign language recognition is a rather new field and many challenges, especially when motion capture devices become more popular. In this paper, we study the feasibility and effectiveness of two classification methods, namely Simplification of Support Vector Machine (SimpSVM) and Relevance Vector Machine (RVM), and also give some comparative results of them for the sign language recognition problem. The experimental results on the Auslan data set and ASLID data set show that SimpSVM and RVM could achieve good predictive performances and SimpSVM is better as compared to RVM on sign recognition. They also pointed out that prediction behaviors of them are similar in terms of the prediction accuracy when the amount of data or the number of feature changed and sign discrimination. However, SimpSVM requires fewer training time than RVM in training phase.