Spatial Constraint-Based Image Relationship Optimization for Large-Scale UAV SfM Reconstruction Using SC-DFS

In large-scale UAV photogrammetry, redundant and weakly correlated image relationships can substantially increase computational costs and reduce matching reliability during SfM reconstruction. To address this issue, this study proposes SC-DFS, a spatial constraint-based image relationship optimization method for efficient UAV image organization and reconstruction. The contribution lies in integrating POS-derived spatial constraints with image relationship graph optimization, where spatial overlap information, image similarity, and F-RANSAC-guided geometric verification are combined to construct weighted image relationships. Subsequently, a depth-first search strategy is employed to extract the connected component containing the largest number of images and optimize the image set for subsequent SfM reconstruction. Experimental results on a large-scale UAV dataset demonstrate that SC-DFS reduces the number of retained images from 2556 to 2274 with visually consistent overall 3D structure. The optimized dataset reduces the sum of the three reported processing-stage times by 2 h and 17 min. The proposed method effectively reduces redundant image relationships, improves image matching efficiency, and provides an efficient solution for large-scale UAV SfM reconstruction in the evaluated acquisition.