Improved Lightweight Human Posture Recognition Based on YOLOv11-Pose

In this paper, an improved lightweight human posture recognition method based on YOLOv11-pose is proposed. By introducing adaptive downsampling module (ADown) and large separable kernel attention mechanism (LSKA), the performance of the model is improved and the computational complexity is reduced. The experiments on MPII-pose data set show that the improved model improves the mAP50 index by 1.65% compared with the benchmark model, reaching 0.676; In the evaluation of mAP50-95, it increased by 6.82% to 0.282; At the same time, the computational complexity (GFLOPs) is reduced by 16.4% and the number of parameters is reduced by 18.6%. This lightweight design significantly improves the running efficiency of the model while maintaining high recognition accuracy, making it more suitable for real-time application in resource-constrained environments.