Design and Implementation of a Deep Learning-Driven Intelligent Volleyball Teaching System

An Intelligent Volleyball Teaching System Driven by Deep Learning allows for automated skill evaluation through the combination of computer vision and advanced learning models which are able to analyze player movements with great precision. Nevertheless, the current volleyball analysis techniques still have some drawbacks like low robustness to changes in camera positions, problems in tracking more than one player at a time, and a limited understanding of intricate temporal motion patterns. In order to address these difficulties, the research suggests a smart framework that combines player detection based on YOLOv8, tracking based on DeepSORT, and a hybrid CNN–LSTM architecture for accurate classification of player actions and assessment of skills. The first step in the proposed methodology is to preprocess the Group-Activity-Recognition-Volleyball dataset which involves frame cleaning, normalization, and data augmentation to improve model generalization. The High-accuracy player detection is performed through the YOLOv8 while DeepSORT tracking ensures the players remain in the same position in the video frames to Kalman filtering and appearance matching. The examination of volleyball actions was performed through several metrics, and it indicated that actions such as l_winpoint (Final Score: 67.51) and moving (Final Score: 65.78) got the best total performance, whereas spiking (49.93) and standing (50.95) scored lower, which is the indication of differences in Accuracy, Timing, Posture, and Arm Swing. These findings point out the benefits of using together biomechanical and performance metrics for thorough action evaluation.