摘要
复杂异型建筑立面很难提取,当前提取方法大都不能达到很高的准确度,为此,提出一种新的基于激光点云扫描技术的复杂异型建筑立面快速提取方法。利用对建筑进行高密度采样,得到复杂异型建筑立面三维空间坐标点云集合,将扫描得到的点云集合拼接纳入统一坐标系。对激光点云数据进行降噪处理,在此基础上,依据点云扫描点的权值设定,求出某网格特征值,针对整个网格系统,形成特征图像。选择数量相同的复杂异型建筑立面和非建筑物立面网格点当成训练样本,对复杂异型建筑立面网格分类模型进行训练,把特征图像当成模型的输入特征,通过线性核函数的SVM获取输出网格,把获取的复杂异型建筑立面网格反映射至三维空间,得到准确的复杂异型建筑立面。将美国某学院的激光扫描点云数据库的三个数据集当成源数据进行实验,发现所提方法准确度与召回率均相对较高,提取的复杂异型建筑立面与实际测量结果相比,倾斜变形以及水平变形均很低,有很高的可靠性。
It is difficult to extract the complex special-shaped building facade,and most of the current extraction methods cannot achieve high accuracy.Therefore,a new rapid extraction method of complex special-shaped building facade based on laser point cloud scanning technology is proposed.Based on high-density sampling of the building,the 3D coordinate point cloud collection of complex special-shaped building facade is obtained,and the scanned point cloud collection is spliced into the unified coordinate system.On the basis of noise reduction of laser point cloud data,the characteristic value of a grid is calculated according to the weight setting of point cloud scanning point,and the characteristic image is formed for the whole grid system.The same number of complex special-shaped building facade and the non-building facade grid points are selected as training samples.The complex special-shaped building facade grid classification model is trained,and the feature image is regarded as the input feature of model to obtain the output grid through the SVM of linear kernel function.The complex special-shaped building facade grid is reflected to threedimensional space to get the accurate complex special-shaped building facade.Three data sets of laser scanning point cloud database of a college in the United States are used as the source data for experiments.It is found that the accuracy and recall rate of the proposed method are relatively high.Compared with actual measurement results,the extracted complex special-shaped building facade has lower tilt deformation and horizontal deformation,and has high reliability.
作者
欧业宁
OU Yening(Beibu Gulf University,Qinzhou Guangxi 535000,China)
出处
《激光杂志》
北大核心
2020年第10期177-181,共5页
Laser Journal
基金
广西壮族自治区自然科学基金项目(No.2018GXNSFBA281129)
广西本科高校特色专业及实验实训教学基地(中心)建设项目(No.桂教高教〔2018〕52号)资助。
关键词
激光点云
扫描技术
复杂异型建筑
立面
提取
laser point cloud
scanning technology
complex special-shaped building
facade
extraction