Sign Language Detection in Video and Real-Time Video Calls

Real-time communication between signers and non- signers remains one of the biggest technological challenges. Most of the current systems use raw video frames, which are prone to a number of factors such as variation in lighting conditions, camera stability, and background clutter and image quality. This paper proposes a novel landmark-based sign language detection system that is designed for both videos and real-time video calls and provides a robust and efficient option against conventional image-based methods. The system integrates Media Pipe land- mark extraction, CNN-LSTM temporal modeling and WebRTC communication, effortlessly decoding both static and dynamic gestures with high accuracy. Representing gestures as numer- ical coordinate sequences rather than raw images, the model shows tremendous adaptability in heterogeneous environments and demonstrates smooth performance during live interactions. Experimental evaluations validate that the proposed architecture ensures reliable detection accuracy, low latency and smooth frame rates, which makes it suitable for realistic communication scenarios