International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 04 | Apr 2025
p-ISSN: 2395-0072
www.irjet.net
Sign Language Vision to Text Using Deep Learning Vinayak Suryavanshi1, Ritika Parab2, Abhijeet Yadav3, Nishek Sharma4, Prof.Vrushali Thakur5 1,2,3,4Student, Dept of Computer Engineering, MGM’s College of Engineering and Technology, Kamothe,
Navi Mumbai, India
5Prof. Vrushali Thakur: Professor, Department of Computer Engineering, MGM’s College of Engineering and
Technology, Kamothe, Navi Mumbai, India ---------------------------------------------------------------------***--------------------------------------------------------------------Traditional methods for sign language recognition have used Abstract - For deaf and hard-of-hearing people, sign
sensor-based gloves, accelerometers, or depth sensors, which are good but sometimes costly, invasive, and impractical for broad use. Some of the pioneering work by Starner and Pentland [2] proved real-time sign language recognition with Hidden Markov Models (HMMs) feasible, setting the stage for current recognition systems. Our system utilizes CNNs to extract visual features and LSTM and Transformer models for recognition of temporal sequences. The solution also exploits the use of MediaPipe to achieve real-time landmark detection as well as the use of OpenCV for preprocessing images and image segmentation. Our objective is to design a low-cost, easy-to-use, high-accuracy system that works within real environments that have unpredictable lighting conditions and backgrounds. This research advances the area of assistive technologies by offering a scalable and versatile system which can be instantiated for various sign languages, is deployable on multiple devices, and can ultimately accommodate features such as sentence recognition and voice synthesis.
language is an essential medium of communication. However, since sign language is not universally understood, it poses a range of problems in social as well as professional contexts. A vision-based deep learning approach using sign language gestures to provide text translation is proposed in this paper, providing better access and more people. In this system Convolutional Neural Networks (CNNs) are used for gesture recognition; and Long Short-Term Memory (LSTM) networks are employed in the temporal sequence learning process. For improved recognition accuracy, the model incorporates attention mechanisms and uses a hybrid feature extraction method. A wide variety of sign languages are used to create the dataset, which covers enough general conditions and a range of scenarios. One must observe that using a variety of representations significantly improves the likelihood of effective generalization. The suggested approach is tested using both standard datasets and custom-collected sign motions. By displaying mistake rates, it produces promising accuracy results. Data augmentation and subs are two adaptive learning strategies that address issues including hand occlusions, signer variability, and various sign language patterns.
Key Words: Sign Language Recognition, Deep Learning, CNN, LSTM, Computer Vision, Gesture Translation, RealTime Processing, Accessibility, Sequence Learning, Human-Computer Interaction.
1.INTRODUCTION Sign language is a rich, expressive visual language that is utilized by millions of deaf and hard-of-hearing people across the globe. Sign languages differ from spoken or written languages in that they are dependent on the accurate combination of hand movement, facial expression, and body movement to communicate meaning. Each nation or region might use its own type of sign language, e.g., American Sign Language (ASL), British Sign Language (BSL), and Indian Sign Language (ISL), each with different grammar and vocabulary [2].
© 2025, IRJET
|
Impact Factor value: 8.315
Fig 1 Sign Gestures 1.1Objectives The primary objective of this study is to create a real-time, vision-based sign language to text translation system based on deep learning. The specific objectives are as follows –
|
ISO 9001:2008 Certified Journal
|
Page 1328