基于空间体素表达与影像目标验证的建筑物三维变化检测分析  

Building 3D Change Detection Analysis Based on Spatial VoxelExpression and Image Target Recognition

作　　者：陈华江 张晋博 张展豪 陈敏 CHEN Huajiang;ZHANG Jinbo;ZHANG Zhanhao;CHEN Min(School of Earth Science and Environmental Engineering,Southwest Jiaotong University,Chengdu 611756,China;China Railway Eryuan Engineering Group Co.,Ltd.,Chengdu 610038,China)

机构地区：[1]西南交通大学地球科学与环境工程学院,四川成都611756 [2]中铁二院工程集团有限责任公司,四川成都610038

出　　处：《测绘与空间地理信息》2024年第5期45-47,51,共4页Geomatics & Spatial Information Technology

基　　金：四川省科技计划(2023NSFSC0247);国家自然科学基金(42371445)资助。

摘　　要：针对复杂城乡区域违法建筑检测效率低、虚警和楼层加盖等漏检多的难题,本文结合三维点云空间体素表达与二维影像目标识别,提出一种适用于无人机影像密集匹配点云的建筑物三维变化检测方法。首先,对两时相的无人机密集匹配点云进行配准、地面点滤波等预处理;然后,对点云构造八叉树,逐层比较体素获取变化点云;最后,将变化点云聚类后逐个按正射视角投影生成二维影像,利用ConvNext网络进行建筑物验证,得到建筑物变化检测结果。本文实验在对象级上进行检查效果评价,在正确率达到87.18%的情况下,实现了100%的检测完整率。结果表明本文方法能够有效提升违章建筑发现效率,满足实际生产需求。Aiming at the problems of low detection efficiency,false alarm and missing detection of illegal buildings in complex urban and rural areas,in this paper,a 3D change detection method for buildings suitable for dense matching point clouds of UAV images is proposed by combining 3D point cloud spacial voxel expression and 2D image target recognition.First,the two-phase UAV dense matching point clouds are pre-processed,including point cloud registration and ground point filtering;then,an octree is constructed for the point cloud,and voxels are compared layer by layer to obtain the changed point cloud.Finally,the changed point cloud is clustered and projected one by one according to the orthographic perspective to generate 2D images.The ConvNext network is used to identify buildings to obtain building change detection results.The experiments evaluate the inspection effect at the object level,achie-ving 100%detection completeness with an accuracy rate of 87.18%.The results show that the proposed method can effectively im-prove the efficiency of illegal building detection and meet actual production needs.

关 键 词：密集匹配点云 违法建筑物 三维变化检测 空间体素 目标识别 

分 类 号：P25[天文地球—测绘科学与技术] TB22[天文地球—大地测量学与测量工程]