Abbreviations and their full names.

Aiming at the problem that it is difficult to accurately calibrate massive Pan-Tilt-Zoom Camera (PTZ) cameras on telecommunication tower and the visualization effect of orthographic geo-image is poor, this paper proposes a new method of realtime orthographic geo-image generating, which is considering Digital Elevation Model (DEM) and semantic information (ROGI-DS). First, through integrating tower cameras with 3D GIS, a camera calibration method based on view fitting (3D GIS-GeoC) is designed. Then, using the trained semantic segmentation model (TCSM), the sky area can automatically be identified and removed. Finally, based on the results of camera calibration and viewshed analysis, and the orthographic geo-image are generated. The results show that: (1) 3D GIS-GeoC method outperforms the traditional Perspective-n-Point (PnP) algorithm;(2) The tower camera semantic segmentation model (TCSM) achieves an accuracy of 96.7%; (3) ROGI-DS method improves the accuracy and visualization of orthographic geo-image under different terrain constraints, and can be used real-time monitoring of natural resources and emergency reliefs.针对通信杆塔上海量云台变焦摄像机（Pan-Tilt-Zoom Camera, PTZ）难以精准标定，且正射地理影像可视化效果欠佳的问题，本文提出一种融合数字高程模型（Digital Elevation Model, DEM）与语义信息的新型实时正射地理影像生成方法（ROGI-DS）。首先，通过将杆塔摄像机与三维地理信息系统（3D GIS）集成，设计了一种基于视图拟合的摄像机标定方法（3D GIS-GeoC）。随后，利用训练完成的语义分割模型（TCSM）自动识别并去除天空区域。最后，结合摄像机标定结果与视域分析结果，生成正射地理影像。实验结果表明：（1）3D GIS-GeoC方法的性能优于传统透视n点（Perspective-n-Point, PnP）算法；（2）杆塔摄像机语义分割模型（TCSM）的准确率达到96.7%；（3）ROGI-DS方法在不同地形约束下提升了正射地理影像的精度与可视化效果，可应用于自然资源实时监测与应急救援场景。