AP-Pose: Enhancing human pose estimation in complex airport scenes with YOLOv8s-Pose and multi-module optimization

Recognizing human behavior in airports is crucial for security management and public safety. To address the issue of low accuracy in human behavior recognition caused by significant lighting changes and complex backgrounds in airport settings, we propose an improved human pose estimation method based on YOLOv8s-Pose, called the AP-Pose algorithm. Firstly, the PSA module is added to the end of the backbone network to enhance the network's ability to model global features, capture the global relationships between different regions in the image, and filter out interference from complex backgrounds. This enables accurate human target localization and pose estimation. Second, the WASP module is introduced into the neck network to improve the spatial capture capability of the pose estimation network, thereby enhancing the localization accuracy of human joint points. Finally, the Adapter module is employed to enrich the feature representation for pose estimation by incorporating information from the original detection task layer and fine-tuning the pose estimation task using this information. Experiments were conducted on a self-constructed airport human pose dataset, and the results show that our improved method significantly enhances pose estimation performance, achieving a detection accuracy of 94.9%, with improvements of 3.3% in precision (P), 3.3% in recall (R), and 3.5% in mAP@0.5 compared to the baseline model. This method enhances pose estimation accuracy and robustness in complex airport scenes while maintaining computational efficiency, achieving a 3.5% mAP@0.5 improvement over YOLOv8s-Pose with a marginal increase in computational complexity (12.5 GFLOPs vs. 10.8 GFLOPs).