A Machine Learning/Deep Learning Hybrid for Augmenting Teacher-LED Online Dance Education


Catherine Hung, Palo Alto Senior High School, USA


For online dancers, learning a dance move properly without the feedback of a live instructor can be challenging because it is difficult to determine whether a move is done correctly. The lack of proper guidance can result in doing a move incorrectly, causing injury. In this work – we explore the use of a hybrid Deep Learning/Machine Learning approach to classify dance moves as structurally correct or incorrect. Given a video clip of the dancer doing a move, such as the grand plie, the algorithm should detect the correctness of the movement. To capture the overall movement, we proposed various methods to process data, starting with deep learning techniques to convert video frames into landmarks. Next, we investigate several approaches to combining landmarks from multiple frames and training machine learning algorithms on the dataset. The distinction between correct and incorrect grand plies achieved accuracies of over 98%.


Deep Learning, Machine Learning, Classification, Online Dance Education.

Full Text  Volume 13, Number 4