Development of an artificial intellegence based performans analysis system using image processing techniques in tennis players

Development of an AI-Based Performance Analysis System for Tennis Athletes Using Image Processing Techniques Aim: This study aims to develop an evaluation system based on artificial intelligence-supported image processing methods to ensure the objective assessment of the technical performance of tennis players. This system aims to analyze tennis movements in detail using the pose estimation approach, which is one of the image processing techniques; at the same time, it seeks to demonstrate the validity and reliability of the system by combining the findings obtained from objective field assessments such as the ITN test. Material and Method: Fifteen athletes who met the participation criteria were included in the study. The research was conducted using a quantitative and experimental method. For this purpose, video footage of tennis players taken from various angles was collected, and these recordings were analyzed through AI-supported deep learning models. Especially the current model named YOLOv8-Pose was preferred. Result: The computer vision-based pose estimation system was developed as part of the research was used to analyze the movement dynamics of tennis players, and the results obtained demonstrated both the technical accuracy and application potential of the model. The system produced satisfactory results in terms of performance metrics and biomechanical evaluations, contributing significantly to the objective analysis of athletes' technical behaviors. The obtained mAP@50 (0,941) and mAP@50–95 (0,772) values ​​demonstrate that the model can detect both simple and complex positions with high accuracy. The average precision (0,867) and recall (0,884) scores reveal that the model both reduces false positive predictions and successfully captures real poses. Conclusion: This study has demonstrated that the YOLOv8-Pose model is a powerful tool for analyzing the movement dynamics of tennis players, both in terms of technical accuracy and practical application. The model has been able to identify complex stroke patterns with high precision, particularly demonstrating successful performance in terms of mAP, precision, and recall values.