Abstract:
As one of the most common sports, tennis also has a high rate of injury-especially among
beginners and previously injured players. General issues such as a torn rotator cuff, tennis elbow,
and pain in the lower back are due to improper techniques and a lack of professional teaching.
Although most research focuses on stroke classification and injury analysis, few models provide
real-time personalized feedback for injury prevention. This project is for the development of a
real-time personalized feedback system of tennis player stroke mechanics.
The current work has been carried out on real-time injury prevention by proposing a hybrid
CNN-GRU model. In this hybrid model, the spatial feature extraction capability of CNN is
combined with that of the GRU network for modeling temporal sequences. Biomechanics are
caught in both 2D and 3D environments through the technologies available for pose estimation
like OpenPose and MediaPipe. The system personalizes feedback based on mapping of this
player-specific data, including age, injury history, and fitness levels, hence making changes in
their techniques while reducing the chance of injury at play.
Performance testing was performed using accuracy, precision, recall, F1 score, and mean squared
error. These demonstrate a high predictive accuracy of substantial reliability, good precision and
recall would serve best for real-time assessment of injury risks. Further work will include the
extension of this model to other sports and injury prevention scenarios.
Computing methodologies -> Machine Learning -> Neural Network -> Convolutional Neural
Networks
Applied computing -> Life and Medical sciences -> Health informatics -> Injury prediction
Human-centered computing -> Ubiquitous and mobile computing -> Sports informatics
Keywords: Tennis Injury Prediction, CNN-LSTM, CNN-GRU, Deep Learning, Pose
Estimation, Real-time Feedback, Convolutional Neural Networks, Long Short-Term Memory
Networks, Sports Biomechanics, Musculoskeletal Injury Prevention, Attention Mechanisms
