Bharatnatyam Pose and Mudra Recognition Using MediaPipe and Deep Features

Bharatnatyam is a traditional dance form from Tamil Nadu. This ancient form of dance, originally known as Sadhir Attam, is well documented in paintings, stone and metal sculptures. There is a wealth of image data pertaining to this dance form and its history, but no general classification models or systems to recognise the dance form as variations in poses exist. The goal of this paper is to recognise and classify bharatnatyam dance forms and poses in real time. Using a customised bharatnataym dataset, we propose a solution that employs the Resnet architecture and Media Pipe to locate landmarks on input frames for hand gestures and poses in order to develop a real-time classification model. The proposed work compares the VGG-16, Resnet, and Mobilenet V2 architectures for real-time classification tasks and concludes that the Resnet model outperforms the others on testing and training data in terms of various evaluation metrics.