Application of Recurrent Neural Network Application in Identifying the Classical Indian Dance Steps from the Video Inputs

In Indian classical dance, posture is crucial because it enables the dancer to retain equilibrium and control while doing the complicated footwork and hand movements. In classical dance, strong posture enables the dancer to communicate the grace and elegance that are distinctive to this age-old classical dance style. The current work aims to create an intelligent system that can recognize various Indian classical dance positions. Bharatnatyam is taken into account for the study because of how difficult it is to perfect the various postures because of how intricate and similar the gestures are. To differentiate between the various moves, experienced supervision is required. Due to their close resemblance and the significant influence of the rhythmic time cycle on the three, three dance postures—Urdhva Hasta Chakra, Urdhva Kona Suchita, and Ardhaalingan—have been discovered. The 72 videos from 24 volunteers are used to accomplish the goal of the current research work. The movies are pre-processed, converted to 720p, and set to 10 frames per second in order to bring uniformity to the dataset. The coordinates of the crucial landmarks are then calculated for each frame retrieved from the pre-processed video dataset using the Python MediaPipe (MP) package. The arctan method is used to determine the angle of the landmarks with respect to the waist using the coordinates of the landmarks. The coordinates also dictate how quickly a dance step is completed. A numpy array that has been trained for the three various dance postures is used to store the angle and speed variables. For the classification, a Recurrent Neural Network (RNN) model is created. The developed model had a 0.7533 accuracy. A Bharatnatyam movie is used to validate the proposed model. The training dance postures can be recognized by the model with accuracy.