Deep Learning-Based Stroke Recognition System for Table Tennis Players

In table tennis, stroke recognition is a significant factor that contributes to the enhancement of a player's ability to analyse their performance and tailor their training program. Given the dynamic nature of table tennis, it is essential to be able to make snap decisions and execute strokes with precision. Algorithms for stroke recognition that rely on manual annotation or established criteria are not always the most resilient or generic alternatives. In this study, the paper provides a new approach to table tennis stroke detection using deep learning. For the purpose of automatically extracting information from raw table tennis video sequences, our method utilises convolutional neural networks (CNNs). Our dataset for training and evaluating the proposed model includes a wide variety of strokes played by both professional and amateur players. Experiments show that our system can reliably identify a variety of table tennis strokes, such as chop, backhand loop, and forehand topspin. In addition, this paper illustrates how our deep learning-based technique outperforms previous methods by doing a comprehensive comparison analysis. Potentially game-changing for table tennis coaching methods, talent evaluation, and performance analysis is the stroke recognition system under consideration. Furthermore, the neural network effectively generalises across the data of a player not included in the training and validation dataset, accurately classifying new strokes with an impressive 99.2% accuracy.