Application of Swin-UNet and pose estimation-based athlete motion quality detection device in IoT systems

With the rapid development of IoT technology, athlete motion quality detection devices based on human pose estimation have significant potential in sports training and health monitoring. This study proposes a motion quality detection device combining Swin-UNet and pose estimation to improve accuracy and robustness. The SwinUNetPose model integrates the Swin Transformer with U-Net to enhance multi-scale information capture and includes a human segmentation module to optimize pose estimation, particularly in complex backgrounds and occlusion scenarios. The experimental results demonstrate that SwinUNetPose achieves superior performance compared to existing methods, excelling in both accuracy (AP) and recall rate (AR) on the COCO dataset, while maintaining a competitive inference speed. The method demonstrates scalability and efficiency, making it suitable for real-time motion quality detection, particularly in multi-athlete scenarios. This study highlights SwinUNetPose’s reliability in advancing sports health monitoring in IoT systems.