Lists
Lists
Maharjan, Sunila, Fernando, Sandra, Shrestha, Subeksha, Mariyanayagam, Dion and Virdee, Bal Singh (2025) Sign language detection and translation using Smart Glove 2.0 using hybrid CNN-transformer model. In: International Conference on Data Analytics and Management (ICDAM 2025), 13-15 June 2025, London (UK).
Smart Glove 2.0 is an improved version of assistive technology developed to improve sign language detection and translation capabilities. This prototype is an evolved version compared to its predecessor, with advanced hardware components like the MPU-6050 GY-25 sensor and the Arduino Nano ESP32 for capturing hand movements and gestures. This work is an upgrade from RNN with LSTM to the hybrid CNN-Transformer model. It was found that the RNN using LSTM, which processes sequential data from flex sensors, was of high accuracy of 85% in our previous research, which was efficient in recognizing finger movements. While the model returned good results, it remained poor at intricate gestures involving a lot of hand movement. In the hybrid CNN-Transformer model, spatial feature extraction and temporal attention mechanisms have been combined; it turns out to be better than the RNN model in terms of its accuracy at 98%. This model has shown better versatility in recognizing a wider range of gestures. Despite these giant strides, numerous challenges remained, including sensor stability and the required computational load. Smart Glove 2.0 poses a strong stride in the translation of sign language; such current innovations will no doubt open avenues for future research aimed at communications that can help the hearing-impaired and hence let in more inclusiveness into mainstream society at large.
![[thumbnail of Sign Language Detection and Translation using Smart Glove 2_0 using Hybrid CNN-Transformer Model.pdf]](https://repository.londonmet.ac.uk/style/images/fileicons/text.png "Sign Language Detection and Translation using Smart Glove 2_0 using Hybrid CNN-Transformer Model.pdf")
Restricted to Repository staff only until 19 October 2026.
Download (1MB) | Request a copy
 |
View Item |