张祥, 杨剑, 吴浩, 王昱人, 郭世泰, 瞿美仙. 利用多尺度张量投票的建筑立面分割方法[J]. 武汉大学学报 ( 信息科学版), 2021, 46(3): 348-352. DOI: 10.13203/j.whugis20180487
引用本文: 张祥, 杨剑, 吴浩, 王昱人, 郭世泰, 瞿美仙. 利用多尺度张量投票的建筑立面分割方法[J]. 武汉大学学报 ( 信息科学版), 2021, 46(3): 348-352. DOI: 10.13203/j.whugis20180487
ZHANG Xiang, YANG Jian, WU Hao, WANG Yuren, GUO Shitai, QU Meixian. Segmentation Method of Building Facade Using Multi-scale Tensor Voting[J]. Geomatics and Information Science of Wuhan University, 2021, 46(3): 348-352. DOI: 10.13203/j.whugis20180487
Citation: ZHANG Xiang, YANG Jian, WU Hao, WANG Yuren, GUO Shitai, QU Meixian. Segmentation Method of Building Facade Using Multi-scale Tensor Voting[J]. Geomatics and Information Science of Wuhan University, 2021, 46(3): 348-352. DOI: 10.13203/j.whugis20180487

利用多尺度张量投票的建筑立面分割方法

Segmentation Method of Building Facade Using Multi-scale Tensor Voting

  • 摘要: 地面三维激光扫描是城市建筑立面数据采集的新方法,由于三维点云具有数据量大、无规则等特点,导致从点云中精确分割城市建筑物信息面临着严峻的挑战。在传统张量投票方法基础上,充分考虑各个尺度下点云立面特征判别的概率,提出了一种多尺度张量投票方法来实现平面分割,更加精确地实现了立面特征的识别。通过实例,将该方法同主成分分析和传统张量投票进行对比与分析,结果表明,多尺度张量投票方法在分割精度方面较优。

     

    Abstract:
      Objectives  The accuracy of plane segmentation using a region-growing algorithm remains an important and challenging topic for terrestrial laser scanning point clouds. The plane segmentation of a region-growing algorithm depends heavily on the seed point, as there are currently no universally valid criteria.
      Methods  Firstly, we applied the tensor voting (including representation, communication, and decomposition) to calculate the plane strength of each input point. During the communication of the tensor voting, we should use varying numbers of votes to obtain plane strengths for different scales. Secondly, we set the threshold for plane feature strength and count the scale that is greater than the threshold. We then calculate the comprehensive plane strength and use it to determine seed points. Then, according to various criteria, including the angle threshold, the distance threshold, and the minimum region size, we used these seed points to perform the plane segmentation.
      Results  On the basis of the traditional tensor voting method, the probability of point cloud feature recognition under various scales is fully considered. A multi-scale tensor voting method is proposed to realize plane segmentation, which realizes more accurate recognition of facade features. Finally, results of contrast experiments with principal component analysis and tensor voting show that the multi-scale tensor voting method is superior in segmentation accuracy.
      Conclusions  For terrestrial laser scanned point clouds, the region-growing algorithm is a classic plane segmentation method and is seriously affected to a large extent by seed points. Although existing research results have shown that the tensor voting has the potential to detect distinct geometric structures, such as points, lines, and planes, determining seed points using tensor voting often is heavily disturbed by effects of scale(i.e., the voten umber). We propose multiscale tensor voting method to improve the selection of seed points in region growing for plane segmentation.

     

/

返回文章
返回