Human Pose Estimation Based on Multi-Angle Augmentation Module

In view of the deficiencies of existing human pose estimation methods in modeling local features in complex pose scenes and expressing final global features, a new human pose estimation framework, the Multi-Angle Modeling Enhancement and Attention Pose Estimation (MAPose), is proposed. By integrating high-resolution features, a feature enhancement module and an attention module, MAPose provides a new human pose estimation method that improves the accuracy of human pose estimation. The Multi-Angle Feature Modeling (MAFM) feature enhancement module improves the feature expression ability in complex pose scenarios by optimizing the global features of the final output from three aspects: local enhancement, multi-scale modeling and dynamic fusion. In addition, we improved the CrossFFN module in the backbone network and introduced a channel attention mechanism, which, in synergy with the MAFM module, significantly improved the model's adaptability to complex poses. The model's average accuracy on the COCO 2017 and MPII datasets was 76.3% and 89.4%, which is a better performance compared to other state-of-the-art models.