基于物方体素约束的深度图融合方法  

Depth map fusion method based on object cube voxel constraint

作　　者：张帅哲 刘振东 刘新[1] 蔡昊琳 屈文虎 张栋 ZHANG Shuaizhe;LIU Zhendong;LIU Xin;CAI Haolin;QU Wenhu;ZHANG Dong(Collegeof Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao Shandong 266590,China;Chinese Academy of Surveying and Mapping,Beijing 100089,China;Tai'an Gold Land Surveying and Mapping Company Limited,Taian Shandong 271018,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590 [2]中国测绘科学研究院,北京100089 [3]泰安金土地测绘整理有限公司,山东泰安271018

出　　处：《北京测绘》2022年第11期1443-1448,共6页Beijing Surveying and Mapping

基　　金：国家基础测绘项目(A2206);中国测绘科学研究院基本科研业务费项目(AR2215)。

摘　　要：稠密重建是多视立体视觉三维重建技术中的关键环节,主要包括密集匹配以及深度融合两步。目前深度融合生成的密集点云仍然存在密度大、质量参差不齐等问题。为此,本文提出一种基于物方体素约束的深度图融合方法。在该方法中,首先根据影像的地面分辨率自适应地确定体素网格尺寸;其次利用场景的几何特征约束计算融合点坐标;然后基于物方体素约束筛选融合点;最后利用无人机采集的滕州市某镇驻地倾斜影像数据以及ETH3D公开数据集进行验证。实验结果表明,本文方法能够有效降低密集点云密度并保证点云质量。Dense reconstruction is a key link in multi-view stereo vision three-dimensional(3D)reconstruction technology,which mainly includes two steps:dense matching and deep fusion.At present,the dense point clouds generated by deep fusion still have problems such as high density and uneven quality.To this end,this paper proposed a depth map fusion method based on object cube voxel constraints.In this method,firstly,the size of the voxel grid was adaptively determined according to the ground resolution of the image;secondly,the coordinates of the fusion point were calculated based on the geometrical feature constraints of the scene;then the fusion point was screened based on the object cube voxel constraints;finally,the tilt image data of a town in Tengzhou collected by unmanned aerial vehicle(UAV)and ETH3D open dataset were used for verification.The experimental results showed that the method proposed in this paper couldeffectively reduce the density of dense point clouds while ensuring the quality of point clouds.

关 键 词：多视立体匹配 深度图融合 物方体素 几何特征约束 权重函数 

分 类 号：P231[天文地球—摄影测量与遥感]