Research on Multi-Person Pose Estimation Algorithms in Complex Motion Scenarios

To address the existing multiplayer pose estimation algorithms, the key point recognition is prone to problems such as omission or misdetection under athlete crossobscuration or complex backgrounds in sports scenes.An efficient detection algorithm based on YOLO11n-pose improvement is proposed. The algorithm replaces the traditional Bottleneck structure in the C3k2 module of the backbone network with the Bottleneck_DCNV2_Dynamic module, which fuses deformable convolution and dynamic convolution mechanisms. By dynamically adjusting the sampling region. The associated occlusion keypoints are predicted based on the visible keypoints. Then the feature enhancement network SDFA is added to the neck network to improve the model’s ability to capture multi-scale features. Finally, the attention module AFGC is introduced to further enhance the extraction ability of features from the key parts of the human body. The experimental results show that the improved model achieves a significant improvement in human keypoint recognition performance in complex motion scenes. On the COCO2017 human keypoint dataset, where the recall, mAP50 and mAP50-95 are improved by $3.5 \%$, $2 \%$ and $3 \%$, respectively.