Hand Gesture Classification for Sign Language Using Artificial Neural Network
Angga Rahagiyanto, Achmad Basuki, Riyanto Sigit, Aditiya Anwar, Moh. Zikky · 2017
The modeling of human hand movements to recognize sign language has been observed by many researchers using various techniques, tools, and types of sensors. The goal is to recognize a powerful, fast and accurate process. One tool that is able to capture full hand movements is Myo Armband. It uses five sensors: Accelerometer, Gyroscope, Orientation, Orientation-Euler and EMG. Each sensor produces different data in scale and size because of clock rate. This condition makes the learning process of hand gesture data becomes more difficult. It happens because of the high dimension data difference. The solutions for handling high-dimensional dataset and for improving the accuracy of dataset is to use feature extraction techniques. A feature extraction process is required to create uniform data. This research uses Moment Invariant method to extract into vector feature of hand gesture. We tested for gesture of alphabet A to Z based on SIBI (Indonesian Sign Language) with static and dynamic movements. There are 26 classes of alphabet and 10 variants with user dependent. We proposed min-max normalization and classification method using Artificial Neural Network (ANN) to classify dataset. The result is produced an accuracy of 93.08% with 1 hidden layer.