Authors : Adhi Dharma Wibawa, Eko Pramunanto and Ifut Rahayuningsih
Abstract: Sign language has been studied by many researchers in the past. However, most of the evaluation techniques were done based on video data, evaluating it using EMG is new approach. Accuracy in performing the sign motion becoming crucial especially when it is needed to be understood by the public. EMG is one of the most advance tools now a days in recording the muscle activity due to a motion. This study was evaluating some motions in Indonesian sign language by using EMG of myoarm sensors. Twenty sign motions were classified by using Naive Bayes to recognize the 20 letters. The 5 participants were involved in this experiment, time domain features such as MAV (Mean Absolute Value), RMS (Root Mean Square), VAR (Variance) and SSI (Simple Square Integral) were used as the input of the classifier. Each motion was repeated 20 times by the participants. The result showed 79% accuracy of the sign language to be recognized. This result was not optimum yet in term of high accuracy. We opinioned that this is due to the variation of motion from one subject when performing one specific task, besides the accuracy in placement the Myo arm sensor’s position. Moreover, variation in duration during performing one motion can also be another issue in accuracy calculation.
Adhi Dharma Wibawa, Eko Pramunanto and Ifut Rahayuningsih, 2019. Sign Language Classification Based on EMG using Time Domain Features Extraction. Journal of Engineering and Applied Sciences, 14: 6274-6281.