A Real-Time Approach for Finger Spelling Interpretation Based on American Sign Language Using Neural Networks

EasyChair Preprint 10746

Abstract

Since old times, sign language has been one of the oldest and most natural forms of communication; nevertheless, because most people do not know sign language, interpreters are in high demand, Based on American Sign Language, we developed a real-time approach for finger spelling that employs neural networks. We will assess the needs and shortcoming of the created sign language system and create a model according to the needs to improve the model accuracy and usage. People who are deaf or dumb rely on sign language interpreters to communicate. However, finding experienced and qualified interpreters for their day-to-day affairs throughout their lifetime is a very difficult task and unaffordable. Data pre- processing and features extraction with gesture recognition has already applied in the past research. However, there is no research on how to create sentences using dynamic gestures and no already created model has the ability to be upgraded to optimize sentences using grammatical structure. The hand is first sent through a filter in our technique, and then it is passed through a classifier, which predicts the class of the hand motions. Using Gaussian method and classification to perfect the image quality to process the image and produce quality results with high accuracy.

Keyphrases: Artificial Neural Networks, Finger Spellings, feature extraction, sign language

@booklet{EasyChair:10746,
  author    = {Muhammad Saif Ur Rehman and Muhammad Rehman Shahid and Iram Shahzadi and Majid Hussain},
  title     = {A Real-Time Approach for Finger Spelling Interpretation Based on American Sign Language Using Neural Networks},
  howpublished = {EasyChair Preprint 10746},
  year      = {EasyChair, 2023}}