Abstract: Deep learning technology is an essential tool for real-time recognition and translation of sign language, allowing for fluid conversations between sign language users. This approach uses deep neural networks to decode camera-captured hand gestures, analysing each movement and translating it into written or spoken words. Deep learning algorithms can recognize sophisticated patterns and discern minute changes in hand forms, gestures, and facial expressions when given a large amount of data. This system, which requires no additional equipment and can identify both static and dynamic signals, promotes inclusive interactions across language boundaries. Labelling ensures uniformity and clarity in gesture detection. TensorFlow's SSD (Single Shot Multibox Detector) technique enhances the speed and utility of real-world interactions by recognizing gestures as full sentences rather than Letters are written individually. This flexible system can detect signals using both American and Indian standards and adjusts differences in background, skin tone, and illumination. The results show exceptional accuracy, with 85% for static motions and 97% for dynamic sequences utilizing LSTM-GRU layers.

Keywords: Deep Learning, Real-time Recognition, Neural Networks, TensorFlow SSD, Gesture Detection, LSTM-GRU Layers

| DOI: 10.17148/IARJSET.2025.124103