摘要
Abstract
Accuracy improvement method of three-dimensional(3D)modeling in unmanned aerial vehicle(UAV)oblique photogrammetry is studied.To address issues such as geometric distortion,texture blurriness,and local holes in complex terrains and occluded environments using traditional methods,a comprehensive accuracy control framework is proposed,which integrates terrain-following flight,multi-source data fusion,and semantic segmentation technologies.Through terrain-following flight technology,the UAV's flight height is dynamically adjusted to ensure consistent image resolution and reduce occlusion.A multi-source data fusion strategy is adopted,combining LiDAR point clouds with oblique images to enhance the geometric accuracy and texture details of the model.Furthermore,the U-Net3D neural network is utilized for semantic segmentation to automatically identify and repair holes and texture mismatches in the model.The experimental results demonstrate that the proposed framework significantly enhances the geometric accuracy and texture quality of the 3D models.The planar mean error is less than 0.05 m,and the elevation mean error is less than 0.08 m,meeting the requirements of 1∶500 scale topographic mapping standards.The effectiveness of multi-source data collaboration and semantical driven restoration on the complex scene modeling is verified.The results of this study can provide theoretical support for the formulation of production standards for real-scene 3D models.关键词
无人机倾斜摄影测量/三维建模/仿地飞行/多源数据融合/语义分割/神经网络/实景三维模型Key words
UAV oblique photogrammetry/3D modeling/terrain-following flight/multi-source data fusion/semantic segmentation/neural network/real-scene 3D model分类
天文与地球科学
