Tennis serve recognition based on bidirectional long- and short-term memory neural networks

The serve is a crucial technique in tennis, providing players the opportunity to master and organize their attacks during competitions. Current tennis serve training methods often lack sophisticated, data-driven tools, and existing recognition techniques rely on single-feature extraction methods focusing on isolated attributes like trajectory or speed. These methods do not fully utilize the comprehensive spatio-temporal information present in video data, resulting in limited accuracy and robustness in recognizing and analyzing different serving techniques. To address these limitations, this paper proposes a tennis serve recognition method using a bi-directional long short-term memory neural network (BiLSTM). Our approach first employs a modified convolutional neural network (CNN) to extract spatial features from images, enhanced by self-attentive weighting to improve feature extraction. It then uses BiLSTM to capture and represent important temporal features, thereby enhancing the model’s ability to recognize and evaluate serving actions. Experimental results demonstrate that our method outperforms existing neural network models in server recognition tasks, effectively addressing the limitations of previous approaches.