KALANETRA — A Mudra Detection Model for Mohiniyattam

This paper presents KALANETRA, a real-time mudra recognition system designed for Mohiniyattam, a prominent classical dance tradition in India. Mudras, the complex hand gestures that are at the heart of this art form, have a lot of meaning, but figuring out how to recognize them automatically has not been a big focus of research. Our system uses a keypoint-based method to get 21 hand landmarks in real time using MediaPipe Hands. It then uses these landmarks to find geometric features like the distances between keypoints, the angles of joints, and the relative orientations of fingers. These features go into a small machine learning classifier that doesn’t need a GPU to work, so it can run on regular consumer hardware. The system achieves 75–80% classification accuracy while running at interactive frame rates after thorough normalization to handle differences in hand size, position, and orientation. KALANETRA has been tested with many users and in different lighting conditions, and it works well where other image-based options fail. This work goes beyond its technical contributions by helping to preserve classical dance heritage in digital form and opening up new ways for interactive, self-guided dance education.