Athlete Action Recognition Algorithm in Table Tennis

We developed an athlete's action recognition algorithm in table tennis, employing deep learning. We applied artificial neural networks (ANNs) and convolutional neural networks (CNNs) in analyzing actions in table tennis. Including system overview, requirements analysis, system workflow, and action recognition module design, feature fusion strategies were established. Four fusion methods were examined: early fusion, late fusion, fully connected layer fusion, and convolutional layer fusion. The proposed algorithm integrated these methods to enhance the accuracy of actions in table tennis. Using the complementary nature of spatial and temporal information, the algorithm optimizes the recognition performance. The algorithm advances action recognition in table tennis and shows the potential of deep learning in sports analytics.