Year: 2025
Type: article
Abstract: Sign language serves as a prevalent means of communication for individuals with hearing and/or speech impairments. AI-driven automatic systems for sign language identification are highly sought after as they can diminish obstacles between individuals and enhance Human-Computer- Interaction (HCI) for the impaired community. The automatic identification of sign language remains a complex challenge due to the intricate structure of sign language in conveying messages. Isolated signs, which refer to individual gestures executed through hand movements, play a crucial role in this process. Over the past decade, research has advanced the automatic identification of isolated sign language from videos by utilizing machine learning techniques. Beginning with a thorough examination of current recognition methods, particularly focusing on available public datasets, the study puts forth an enhanced convolution-based hybrid Inception architecture aimed at increasing the recognition precision of isolated signs. The primary contributions are the enhancement of InceptionV4 through optimized back-propagation across uniform connections. Furthermore, an ensemble learning framework incorporating various Convolution Neural Networks has also been introduced and leveraged to further boost the recognition accuracy and resilience of isolated sign language recognition systems. The efficacy of the proposed learning methods has been validated on a benchmark dataset consisting of isolated sign language gestures. The experimental findings illustrate that the proposed ensemble model surpasses sign identification performance, achieving a higher recognition accuracy (98. 46%) and enhanced robustness.
Authors: Yarramsetty Sandeep -, Dr Shanti S -
Language: English
Subfield: Human-Computer Interaction
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4410430565