Penerapan Algoritma Gaussian Blur dalam Sistem Penerjemah Bahasa Isyarat

Abstract

In 2021, it was estimated that around 1.3 billion people or approximately 16% of the world's population experienced various types of disabilities including sensory disabilities such as deafness and muteness. Although sign language is used as a non-verbal communication tool. People with unfamiliar Sign language movements are often difficult to understand it. Sign language interpreter services have increased in recent years but still have limitations. Deep Learning technology, particularly Convolutional Neural Networks (CNN) has shown advancements in image and video recognition. CNN is an effective architecture for object classification and has been used to enhance motion detection in hand gesture recognition. The Gaussian Blur technique is used to reduce visual noise in images thereby improving focus on hand gesture recognition in sign language. This research develops a sign language interpreter system that applies background blur. The Tensorflow.js and BodyPix libraries can be utilized to implement this technique by separating body parts from the background allowing the blur to be applied to the background without affecting the body parts. The dataset used in this research consists of 3,480 images of sign language movements, 2,420 images for training data and 1,040 for validation data with 20 categories/classes. The model used is Convolutional Neural Network which was trained using this dataset achieve model accuracy of 91%. This model used to classify sign language movements into words or sentences.