Badminton Action Analysis Using LSTM

Badminton is a popular racquet sport that involves quick movements, agility, and reflexes. A badminton action analysis is a process in which quantify the various actions performed by the player using various sensors or image processing-based approaches. In this paper, a sensor-based approach to collect player action data for various shots, analyse the same and create a Long short-term memory-based model to classify various shots. The trained Long short-term memory model has a Micro F1 score of 0.84. The Model would then generate a shot accuracy score between 1-10. A professional badminton coach also rated the shots, and a comparison was made between the score given by the coach and the ML model. Additionally, arm section-wise comparison of the shot was made. This allows for a more granular analysis of the shot. A comparison of the arm-wise shot is also conducted. It can be seen from the results that the scores given by the coach and the Long short-term memory-based ML model matched by 98.6%. So it can be concluded that such a sensor-based wearable badminton teaching technique can be deployed to enable underprivileged beginner badminton players to achieve higher shot accuracy by allowing them to correct postures during the swing action of the arm.