Smart fitness with YOLO-Fit IoT: Real-time pose analysis and personalized training via IoT and RL

With the development of smart technology, smart fitness is becoming an integral part of modern fitness. Traditional fitness methods depend on coaches or predetermined training plans, often lacking real-time feedback and personalization, which limits their ability to meet the evolving needs of users. Although current smart fitness systems combine various technologies, they still struggle with issues such as inaccurate data fusion, delayed real-time feedback, and insufficient personalization. To address these challenges, this paper proposes the YOLO-Fit IoT model, which integrates IoT technology, YOLO-based posture estimation, and reinforcement learning for real-time exercise posture analysis and personalized training recommendations. By fusing multimodal data, the model processes physiological data and exercise status in real time, dynamically adjusting the training plan based on user feedback. Experimental results, compared to baseline models such as MediaPipe Pose, MS-G3D, FitLSTM, YOLOv7-Pose, and YOLOv10, show that YOLO-Fit IoT significantly outperforms existing systems, offering improvements in accuracy, real-time feedback, and training effectiveness. The combination of IoT devices, deep learning, and reinforcement learning allows for personalized, dynamically optimized training, leading to enhanced fitness outcomes. This research presents a new solution for smart fitness systems, with potential for future optimization in computational efficiency and application expansion to various exercise scenarios.