Human Pose Estimation Method Based on GRSA-YOLOv8n Pose

An efficient human pose estimation algorithm GRSA-YOLOv8 Pose is proposed based on the YOLOv8n Pose. A GELAN module is integrated to enhance the receptive field, the RFA mechanism is combined to capture long-distance context information, and the GSConv convolution is utilized to reduce computational redundancy. A self-attention mechanism is also embedded into the head network for feature extraction. Experimental results demonstrate that the improved model reduces parameter count and computational complexity, achieving a 1.4% increase in mAP50, a 2.4% increase in mAP50:95, a 1.2% improvement in precision, and a 2% boost in recall, with the detection performance significantly enhanced.