基于改进YOLOv11-Pose的实时多人姿态估计算法OA

In view of the insufficient robustness in keypoint localization and low efficiency in feature fusion for 2D human pose estimation in complex scenarios,this paper proposes MCM-YOLO Person,a real-time multi-person pose estimation algorithm based on improved YOLOv11-Pose.A weight-reused multi-scale dilated convolution attention residual(MSDAR)module is designed to effectively extract local features of human keypoints,which significantly improves localization accuracy.An improved cross-context dynamic fusion(CCDF)module is designed to effectively enhance the robustness of keypoint detection in scenarios involving occlusion and complex poses.Additionally,a multi-feature dynamic guided fusion(MFDGF)module is constructed to achieve efficient multi-scale feature fusion,which strengthens the model adaptability to complex human structures and pose variations.Experiments on the MS COCO 2017 keypoint detection dataset demonstrate that the proposed model achieves a 2.1%improvement in average precision(AP)over the baseline YOLOv11-Pose,without compromising inference efficiency.The proposed strategy is also effective on a larger model,attaining an AP of 71.2%and outperforming most mainstream methods.