A Sustainable Aerial Photogrammetry System for Cultural Landscape Modeling using Enhanced Multi-View Stereo

To address the real-world challenges of complex terrain, dense buildings, and severe occlusion in 3D modeling of traditional villages, this study proposes an optimized multi-view stereo reconstruction algorithm based on unmanned aerial vehicles (UAVs). This algorithm improves the integrity, accuracy, and occlusion robustness of the model reconstruction. Based on a stable multi-rotor unmanned aerial vehicle (UAV), the algorithm integrates GNSS/IMU navigation, a three-axis gimbal, and multi-sensor data to achieve high-quality image acquisition at the hardware level. To cope with complex imaging environments, the study designed a multi-target route planning and visual navigation fusion mechanism. This mechanism generates adaptive trajectories by constraining image overlap, shooting angle, and path length. In the image processing stage, the model constructs a multi-scale cost volume and introduces depth hypothesis modeling. This model embeds collinear geometric constraints and an attention mechanism in regions with weak texture and occlusion. Furthermore, in the post-reconstruction processing stage, the model optimizes pose estimation based on image reprojection errors and introduces a multi-scale hierarchical modeling strategy for the first time. The experimental part selected a self-collected traditional Huizhou village image set and a public wooden structure building dataset for verification. In terms of the completeness index, the algorithm achieved 91.7% and 89.3% respectively; the point cloud density increased to 639 pts/m²; the average error of the occluded area was reduced from 12.3mm to 5.1mm, and the number of complete facets in the occluded area of the building facade increased to 589.