3D CNN Approach for Tennis Movement Recognition Using Spatiotemporal Features of Video

The primary goal of this investigation is to develop a model for recognizing tennis movements using the spatiotemporal features of the video data. The study collects and organizes relevant video data, employing machine learning algorithms to create and train neural network models for movement classification. The implementation utilizes a 3D Convolutional Neural Network (3D CNN) to classify tennis player movements effectively. The experiments included a comparison of two variations of the created dataset and models that received different numbers of frames as input. The results showed that the best model achieved an overall accuracy of 64% for 7 types of movements and an accuracy of over 70% for the movements of serve, backhand, backhand slice, backhand volley, and smash. Also, based on the results, it can be concluded that the use of 3D models can show good results and that it is worth continuing to experiment with their different types.