HR-YOLOv8-DE: Advancing Human-Robot Interaction with Precision Pose Estimation in Sports Rehabilitation

Abstract

The integration of human-robot interaction within sports rehabilitation marks a significant advancement in improving the precision and effectiveness of therapeutic exercises. However, existing models often fail to accurately capture and analyze the intricate, dynamic movements required in rehabilitation, resulting in inadequate feedback and less optimal patient outcomes. To address these challenges, we propose the HR-YOLOv8-DE network, which integrates Diverse Branch Block (DBB), HRNet, and Efficient Multi-Scale Attention (EMA). This model is designed to enhance multi-scale feature extraction, maintain high-resolution pose estimation, and adaptively prioritize relevant features across varying conditions. Experimental results demonstrate the superior performance of our approach, with the HR-YOLOv8-DE model achieving a PCK of 82.0% on the MPII dataset and a mAP of 74.7% on the COCO dataset, significantly outperforming existing methods. These advancements not only improve the accuracy and adaptability of human motion analysis in rehabilitation but also set a new standard for future developments in robotic-assisted therapeutic interventions.