Deep Learning-based Classification for Gesture Detection using Xception Model

Gesture language functions as a communication medium for individuals who are dumb and deaf, utilizing recognized signs or bodily movements to convey messages. It incorporates various elements such as shapes, hand motions, directions, and facial expressions. Each sign not only represents a word but also conveys a specific tone. For many deaf individuals, verbal communication is not feasible, and they may encounter difficulties in reading and writing. Thus, the creation of a sign language translation system, specifically a sign language recognition (SLR) system, holds significant importance in their lives. To address this challenge, we utilized the Xception Model. Our approach involved training the Xception Model using a dataset representing the ASL Alphabet, which comprises 24 classes corresponding to English sign language letters (excluding J and Z, which involve motion). The training dataset consisted of 27,455 instances, while the test dataset contained 7,172 instances. The results of our proposed model were highly promising, achieving an impressive accuracy rate of 99.9% with a validation accuracy of 100%. This success highlights the potential effectiveness of our model in automatically recognizing American Sign Language gestures, thus improving Interaction and accessibility for people with hearing impairments.