A Multi-Criterion Pose Estimation Algorithm Based on Improved YOLOv11-Pose

With increasing health risks among the elderly, accurate fall detection has become a global priority. Aiming at the problems of single fall-judgment method, poor real-time performance, and inability to determine the fall direction in current fall detection algorithms. This paper presents an improved YOLOv11-Pose framework, replacing the backbone with MobileNetV4 and adding Shuffle Attention to enhance features. A multi-rule scoring system based on skeletal keypoints is used to estimate posture. Experiments demonstrate that the proposed algorithm can reliably distinguish postures and accurately identify fall directions, providing valuable support for fall risk assessment.