利用凝聚层次聚类的多视影像重建算法  

A Reconstruction Algorithm of Multi-View Images Using Agglomerative Hierarchical Clustering

作　　者：卢俊[1] 张保明[1] 欧小平 滕飞 

机构地区：[1]信息工程大学 [2]测绘信息中心 [3]61243部队

出　　处：《测绘科学技术学报》2015年第2期157-163,共7页Journal of Geomatics Science and Technology

基　　金：国家973计划项目(2012CB720000)

摘　　要：无序多视影像的三维重建由于缺乏先验信息而可能导致重建结果发生漂移。提出一种稳健的重建算法,以匹配点对数量作为影像间的相似性量度,利用层次聚类算法对影像集进行分层分组,通过自下而上的顺序,对各层中的每一组子影像集进行重建;然后利用公共点将相邻子影像集间的重建结果进行融合,直到完成整个影像集的重建。最后通过3组具有代表性的影像集的重建试验,验证了算法的有效性。Three-dimensional reconstruction of unordered multi-view images may lead to drift due to the lack of prior information. A robust reconstruction algorithm is proposed, the image set is hierarchical grouped using agglomerative clustering algorithm in which the number of the matching points is used as the affinity measure between images, and every sub image set in each layer is reconstructed through the bottom-up order, and then the reconstructions of the adjacent sub image sets are merged through common 3D points, until the completion of the reconstruction of the entire image set. Finally, the reconstruction experiment of three representative image sets verifies the effectiveness of the proposed algorithm.

关 键 词：增量式重建 并查集 凝聚层次聚类 树状图 分层重建 

分 类 号：P237[天文地球—摄影测量与遥感]