Indonesian Sign Language (BISINDO) Recognition Using Accurate and Fast Dynamic Time Warping Learning Model
Tri Handhika · International Journal of Machine Learning and Computing · 2020
Sign language recognition problem should be represented as a time series classification model with high accuracy. In the previous studies, Indonesian sign language (BISINDO) had been modeled with one of stochastic time series classification model, i.e. Hidden-Markov Model (HMM), but has low accuracy. In other studies, BISINDO had been recognized with high accuracy but using an unrepresentative model (non-time series classification model), i.e. the modified Generalized Linear Vector Quantization (mGLVQ) model with mode function. In this paper, we tried to use a deterministic time series classification model, named Accurate and Fast Dynamic Time Warping (AF-DTW) model. AF-DTW model is a modified form of Dynamic Time Warping (DTW) model. It is not only to improve the accuracy of DTW but also to accelerate the finding of optimal warping path. The output results showed that AF-DTW has a much higher accuracy than HMM, although it is not as accurate as mGLVQ.