一种改进顶帽变换与LBP高程纹理的城区建筑物提取算法

利用LiDAR数据的建筑物提取存在植被点与建筑物点难以区分的问题,利用航空影像进行城区建筑物提取则无法有效剔除阴影区域植被。本文融合LiDAR和航空影像两种数据源,提出了改进顶帽变换及局部二进制模式(LBP)高程纹理分析的建筑物提取算法。首先将LiDAR数据进行规则格网化,通过改进顶帽变换提取地面数据点,然后根据航空影像计算归一化差值植被指数(NDVI)值进行植被粗提取,计算LBP高程纹理,精细区分植被点与建筑物点,最后利用形态学操作填充建筑物孔洞,以检测出的建筑物点为种子点进行区域生长,得到完整的建筑物点集合。试验基于ISPRS提供的Vaihingen数据集中复杂多植被城区场景,试验结果表明,本文算法能够有效区分植被与建筑物,实现建筑物准确提取。

An Algorithm of Building Extraction in Urban Area Based on Improved Top-hat Transformations and LBP Elevation Texture

Classification of building and vegetation is difficult solely by LiDAR data and vegetation in shadows can't be eliminated only by aeri al images .The improved top-hat transformations and local binary patterns (LBP) elevation texture analysis for building extraction are proposed based on the fusion of aerial images and LiDAR data .Firstly ,LiDAR data is reorganized into grid cell ,the algorithm removes ground points through top-hat transform .Then , the vegetation points are extracted by normalized difference vegetation index (NDVI) .Thirdly ,according to the elevation information of LiDAR points ,LBP elevation texture is calculated and achieving precise elimination of vegetation in shadows or surrounding to the buildings .At last ,morphological operations are used to fill the holes of building roofs ,and region growing for complete building edges .The simulation is based on the complex urban area in Vaihingen benchmark provided by ISPRS ,the results show that the algorithm affording higher classification accuracy .