GROMS: GPS-free real-time UAV orthophoto mapping via terrain constraint monocular SLAM

To improve the coverage of individual Unmanned Aerial Vehicle (UAV) images and enable comprehensive environmental perception, it is essential to generate wide-field, large-scale, and metrically accurate orthophotos. However, traditional methods depend heavily on the Global Positioning System (GPS), exhibit high computational complexity, and are susceptible to drift and distortion in GPS-free environments. To overcome these challenges, we propose GROMS, a GPS-free, map-prior-assisted (DEM and satellite imagery) real-time UAV orthophoto mapping system. GROMS utilizes georeferenced satellite imagery and digital elevation model (DEM) solely as map priors to derive virtual control points (VCPs) and constrain monocular SLAM, without relying on any GNSS or IMU measurements. Specifically, we design an efficient keyframe selection and dynamic block matching module to enable robust alignment between UAV and satellite images. We further introduce a terrain-constrained monocular SLAM framework incorporating VCPs, enabling drift-free and high-precision pose estimation in the absence of GPS/IMU. Finally, a hybrid ortho-rectification and incremental multi-band blending strategy is proposed for seamless, real-time orthophoto generation. Extensive experiments in both typical and challenging scenarios validate the robustness, accuracy, efficiency, and visual fidelity of GROMS. The system achieves visual quality comparable to commercial software, with efficiency improvements of up to 7.6× . Under GPS-free conditions, GROMS attains mean absolute errors of 1.11 m and 0.64 m in the x- and y-directions, respectively, while sustaining stable end-to-end throughput with bounded latency (p99 < 2.8 s) during continuous online orthophoto generation. Demonstration videos of GROMS and the associated datasets are available at https://github.com/YFS90/Real-time-orthophoto .