摘要
针对无人机影像获取密集点云和DSM过程中耗时过多、点云质量与处理速度难以平衡的问题,本文提出了一种大场景无人机在线密集点云和DSM生成方法。首先,采用SLAM与RTK信息松耦合的方法在线估计影像位姿;然后,利用多视影像位姿信息在深度空间进行离散采样,通过将当前像素点离散采样深度投影至候选帧影像上获取匹配代价,在代价计算过程中引入中心对称的census代价函数,并考虑多视图遮挡关系计算联合代价值,节省时间的同时提高了匹配代价准确度;最后,提出了一种增量式的代价积聚策略,将前一帧获取的深度投影至当前帧约束代价积聚范围,从而缩短在线计算耗时,结合抛物线拟合算法得到完整度和精度更高的深度图,将深度图去噪后投影至物方空间得到最终的密集点云和DSM。利用3组典型地区的无人机影像对本文方法进行测试,结果表明,本文方法能够满足在线计算的要求,获取的点云和DSM精度与完整度较好。
Aiming at the problem that UAV image acquisition of dense point cloud and DSM is taking too much time and point cloud quality and processing speed are difficult to balance,a large scale online UAV mapping algorithm for the dense point cloud and digital surface model generation is proposed.Firstly,the loose coupling between SLAM and RTK information is used to estimate the image pose online.Then,we use multi-view image pose information in depth space discrete sampling,discrete sampling depth through the current pixel are projected to candidate for getting the matching cost.We use a symmetrical census cost function to compute cost volume,and considering multiple view shade relations joint generation value calculation to save time and improve the accuracy of matching cost.Finally,we propose an incremental cost accumulation strategy to constraint depth range on the current frame using the previous frame depth,to decrease online computation time consuming,and combined with a parabolic curve fitting method to get the higher accuracy and completeness of final depth,the final dense point cloud and DSM results are obtained by projecting the depth map to the object space after drying.The experimental results of three groups of UAV images in typical areas show that our method not only can satisfy the requirements of online computing,but also the accuracy and completeness of point cloud and DSM result are good.
作者
杨佳琪
范大昭
杨佳宾
杨幸彬
纪松
YANG Jiaqi;FAN Dazhao;YANG Jiabin;YANG Xingbin;JI Song(Institue of Geospatial Information,Information Engineering University,Zhengzhou 450001,China;Beijing Zitiao Network Technology Co.,Ltd.,Beijing 100086,China)
出处
《测绘通报》
CSCD
北大核心
2023年第10期47-53,共7页
Bulletin of Surveying and Mapping
基金
高分辨率对地观测系统重大专项(42-Y30B04-9001-19/21)
国家自然科学基金(41971427)。
关键词
无人机影像
匹配代价计算
增量式代价积聚
密集点云
在线计算
UAV image
matching cost calculation
incremental cost accumulation
dense point cloud
online computation
