A Lightweight Human Pose Estimation Algorithm Based on Improved YOLO11-Pose

As computer vision technology is applied to human pose estimation, current human pose estimation models suffer from problems such as large computational load and parameter count and slow inference speed. This paper proposes an improved lightweight human pose estimation algorithm based on YOLO11-Pose. By introducing Dysample dynamic upsampling in the Neck section to replace nearest neighbor interpolation upsampling, the accuracy of human body keypoint recognition is improved; RCSPELAN is proposed, and the C3k2 module in the model is replaced to enhance feature extraction and reduce computation and parameter requirements; meanwhile, a DESD detection head is proposed by employing detail-enhancement convolution, which effectively captures key details of the human body and reduces model parameters and computational complexity. On the MS COCO human keypoint dataset, mAP50 increased by 0.3%. At the same time, the number of model parameters decreased by 29%, and the computational cost dropped by 13.5%, resulting in a lightweight model. Received: 9 November 2025 | Revised: 6 February 2026 | Accepted: 27 February 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement This paper utilizes the MS COCO human keypoint dataset, which is publicly available at https://cocodataset.org/#download. Author Contribution Statement Yang Gao: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Project administration. Guanglei Qiang: Methodology, Software, Validation, Formal analysis, Investigation, Resources, Writing – original draft, Visualization, Supervision, Project administration. Fujiang Yuan: Methodology, Software, Investigation, Resources, Data curation, Writing – review & editing, Visualization, Supervision.