中山大学学报（自然科学版）（中英文）2026，Vol.65Issue(1)：76-84,9.DOI:10.11714/acta.snus.ZR20250069
面向消防救援指挥的无人机航拍快速成图方法
A rapid mapping method for UAV aerial photography in fire rescue
摘要
Abstract
This study proposes a semantic information-based aerial image stitching methodology.In the first stage,a coarse registration method using projection transformation based on 3D vector rotation is adopted.Through geographic coordinate conversion,view frustum modeling,plane intersection calculation,and projection matrix estimation,efficient coarse registration of aerial images onto a unified elevation plane is achieved.In the second stage,a semantic segmentation and LightGlue feature based optimization method is utilized.Each row of images after coarse registration is treated as an independent optimization unit:road topological structures are obtained via semantic segmentation,and high-precision registration is realized by combining a lightweight feature matching network with geometric constraint optimization.Experimental results show that the proposed two-stage optimization model effectively overcomes the dependence of camera intrinsics.Finally,the accurate disaster-stricken locations are visualized on the geographic system interface,enabling comprehensive and systematic recording of disaster characteristics.关键词
消防救援/无人机/正射图像/语义分割Key words
fire rescue/unmanned aerial vehicle/orthophoto/semantic segmentation分类
信息技术与安全科学引用本文复制引用
卜祥鹏,赵涂昊,翟浩州,胡天江..面向消防救援指挥的无人机航拍快速成图方法[J].中山大学学报（自然科学版）（中英文）,2026,65(1):76-84,9.基金项目
广东省重点领域研发计划(2024B1111060004) （2024B1111060004）
