Advancing Indian sign language recognition: a comparative deep learning study using YOLOv11, Vision Transformers, and custom CNN architectures

Indian Sign Language (ISL) recognition plays a pivotal role in bridging the communication gap between the hearing-impaired and the general population, fostering inclusive and accessible communication systems. This study presents a comprehensive comparative analysis of three state-of-the-art deep learning approaches—YOL0v11, Vision Transformers (ViT), and a Convolutional neural network (CNN) developed from scratch—for ISL gesture prediction. The models were trained and evaluated on a curated ISL dataset, with performance assessed in terms of classification accuracy and loss across five training epochs. The YOL0v11 model achieved a flawless 100% accuracy, demonstrating exceptional real-time detection capabilities. The ViT model achieved a rapid convergence, reaching 100% accuracy by the second epoch with a final loss of 0.0513, highlighting its effectiveness in capturing spatial and contextual dependencies. The custom CNN model delivered competitive results, attaining 99.89% validation accuracy with minimal validation loss (0.0041) after five epochs. These findings underscore the viability of deep learning-based solutions for ISL recognition, with YOL0v11 emerging as the optimal choice for real-time deployment, while ViT and CNN architectures offer robust alternatives for classification tasks. This work provides a strong foundation for developing intelligent assistive technologies to enable seamless interaction with hearing-impaired individuals in multilingual contexts..