DeepMinton: Analyzing Stance and Stroke to Rank Badminton Players

Abstract

In recent times, wearable devices have gained immense popularity for various pervasive computing and Internet-of-Things (IoT) applications including sports analytics. Recent works in sports analytics primarily focuses on improving a player's performance and help devise a winning strategy based on the player's strengths and weaknesses. In a racquet-based sport, it is often perceived that the way the racquet is being posed mostly influences the performance of the players. However, in this work we posit that the stance and the posture of the players are also of equal importance. Indeed a perfect posture and stance allows a player not only to play a stroke efficiently by directing the shuttle to strategic spots but also making it difficult for the opponent to return the shot and score a point. Therefore, we hypothesize that the performance of a player equally correlates with the stance and the efficiency of handling the racquet. In this theses, we propose DeepMinton, a data-driven framework to analyze the stance and posture of the badminton players based on the different shots that are played and rank them based on their performances. First, we employ machine learning algorithms to classify the strokes and stances of the players. Second, we propose a distance-based methodology to compare the stances of an intermediate and a novice player with that of a professional player. Third, we quantify the error between the professional player'sstance with that of an intermediate and a novice player. Finally, we devise a deep convolutional regressor to predict the score of a shot, that helps rank the players based on their performances. We evaluate DeepMinton at Badminton courts in UMBC RAC (Retrievers Activities Center) using 4 Shimmer3 IMU Unit devices comprising of accelerometer sensors by placing on the dominant wrist, palm, and both the legs of the players. We collected the data from a novice player, an intermediate player and two professional players for 12 different frequently played shots. Empirical results indicate that DeepMinton achieves 89.09% accuracy for strokes classification and 79.91% accuracy to detect stance errors.