<p>As a significant visual processing technology in IoT, human pose estimation attract much attention recently and has been widely used in remote pedestrian monitoring systems within wireless environment. To solve the occlusion problem in pose estimation, this paper introduces a graph structure-aware human pose explicit modeling network. This network is divided into two parts: the backbone network and the graph structure-aware layer built using spatial graph convolutional networks. Considering the adjacency matrix of keypoints is an important input for graph neural networks, this study designs a connection method based on keypoint clustering. Spectral clustering of the keypoint correlation matrix is performed to determine the keypoint adjacency matrix using statistical machine learning methods, addressing the impact of complex poses. By stacking graph structure-aware layers and utilizing the connection information between keypoints, the strong correlation between keypoints is explicitly modeled to infer the location of invisible keypoints and refine the preliminary pose output by the backbone network. Through quantitative and qualitative analysis, this study proves that the graph structure-aware human pose explicit modeling network can effectively address occlusion problems and correct the position of invisible keypoints.</p>