Dynamic Indonesian sign language recognition by using weighted K-Nearest Neighbor

People with disability in Indonesia is 2.45% of the total population. And people with hearing disability are the second largest. Usually, they have a problem with communication. A sign language is created to solve the problem. However, it is less popular so only limited people understand it. Therefore, a system which able to recognise sign language is required. This research introduces dynamic feature extraction to recognise sign language. Dynamic features were used to capture trajectory movement of hand skeleton. Because sign languages are characterized not only by hand movement but also by hand position, besides dynamic features this research also used hand position feature. For classification, this study used a combination of Weighted Simple Matching Coefficient (WsMC) and K-Nearest Neighbors (KNN). Later combination between Weighted Simple Matching Coefficient (WSMC) and K-Nearest Neighbors (KNN) is called by Weighted K-NNN. For the experiment, eighteen sign language gestures were tested. Tree kinds of K-NN was applied in testing; K-NN; locally weighted K-NN; and globally weighted K-NN. The recognition accuracy of this system, although evaluated with a limited vocabulary, presents very promising result with value 88% by using locally weighted K-NN.