Recognition method of tennis swing based on time series convolution network
Bo Huang · Molecular & cellular biomechanics · 2025
Tennis swings vary widely in type, and accurately identifying these motion patterns is crucial for swing analysis. With advancements in artificial intelligence, recent studies have achieved significant progress in human activity recognition through machine learning and sensor technologies. However, research specifically on tennis swing recognition remains relatively nascent, with limited exploration in this domain. This study focuses on recognizing tennis swing motions using a time-series convolution network, employing sensors to gather essential motion data. The MPU9250 sensor captures the intricate nuances of human movement, which often displays complexity and individual variation. Key challenges include effectively extracting features of tennis swings, designing suitable classifiers for recognition, and enhancing classifier generalization across different individuals. Addressing these challenges, this study introduces a temporal sensing network for swing recognition based on causal and dilated convolution techniques. The network effectively captures the temporal characteristics of swings, achieving a 94.73% recognition rate. Additionally, a comparative analysis between the sequential convolution network and traditional machine learning algorithms is conducted, providing insights into their performance and processing workflows.