测绘科学技术学报2025，Vol.41Issue(4)：372-380,9.DOI:10.3969/j.issn.1673-6338.2025.04.006
融合语义分布与神经辐射场的无人机图像三维表面重建方法
A Method for 3D Surface Reconstruction of UAV Images Integrating Semantic Distribution with Neural Radiance Fields
摘要
Abstract
To address the issue of insufficient coupling between geometric information and radiance information in existing three-dimensional(3D)reconstruction methods,which leads to the reconstruction outcomes being heavily influenced by dynamic objects and shadows,a novel method for 3D surface reconstruction from unmanned aerial vehicle(UAV)images is proposed,which integrates semantic distribution with neural radiance fields.By embed-ding categorical information,the semantic distribution is incorporated into the neural radiance field model,allowing it to be optimized alongside the radiance information.This approach facilitates a positive interaction be-tween semantic and imaging characteristics,and realizes the geometric-semantic integration processing for high-fi-delity 3D reconstruction and object attribute information extraction.The experimental results demonstrate that the incorporation of semantic distribution leads to more coherent light state reconstruction in the neural radiance field model.This improvement effectively enhances performance across various tasks,including novel view rendering,object attribute detection,and digital surface model(DSM)reconstruction.Consequently,the proposed method not only improves 3D surface reconstruction but also better supports downstream tasks such as dynamic object detection and shadow recognition.关键词
无人机图像/神经辐射场/语义分割/上下文信息/三维表面重建Key words
drone images/neural radiance field(NeRF)/semantic segmentation/contextual information/3D sur-face reconstruction分类
测绘与仪器引用本文复制引用
李力,张永生,江志鹏,王自全,高寒,于英..融合语义分布与神经辐射场的无人机图像三维表面重建方法[J].测绘科学技术学报,2025,41(4):372-380,9.基金项目
国家自然科学基金项目(42071340) （42071340）
嵩山实验室项目(221100211000-01 （）
221100211000-04). （）
