Real-Time Sign Language Recognition using MediaPipe Landmark Pipeline and Deep Learning Classifiers

Communication between normal individuals and people with hearing and speech disabilities is often difficult due to the lack of an effective interaction method. This paper proposes a real-time sign language recognition system using a MediaPipe landmark pipeline and deep learning classifiers. The system uses a standard webcam to capture video frames and extracts 21 hand landmarks using MediaPipe Hands. These landmarks are converted into feature vectors and processed by a dual-classifier model consisting of a Multilayer Perceptron (MLP) for static gestures and a Long Short-Term Memory (LSTM) network for dynamic gestures. The proposed approach provides accurate and efficient real-time recognition of sign language gestures. Experimental evaluation on ASL and ISL datasets shows that the system achieves high accuracy and low latency, making it suitable for real-time communication applications.