Oblique View Selection for Efficient and Accurate Building Reconstruction in Rural Areas Using Large-Scale UAV Images

3D building models are widely used in many applications. The traditional image-based reconstruction pipeline without using semantic information is inefficient for rural areas. An oblique view selection methodology efficient and accurate areas proposed this paper. A Mask R-CNN model trained satellite datasets to detect instances nadir UAV images. Then, the detected images directly georeferenced. georeferenced select that cover buildings by nearest neighbours search. Finally, precise match pairs generated from selected their principal points. tested on a dataset containing 9775 total of 4441 covering 99.4% all survey area automatically selected. Experimental results show average precision recall 0.90 0.88, respectively. percentage robustly matched oblique-oblique oblique-nadir image above 94% 84.0%, evaluated sparse dense reconstruction. based reduces 68.9% data processing time, it comparably complete. also high consistency between point clouds reconstructed methodology.