多视密集匹配并行传播GPU-PatchMatch算法  

A GPU-PatchMatch multi-view dense matching algorithm based on parallel propagation

作　　者：邓非[1,2] 颜青松 肖腾 DENG Fei;YAN Qingsong;XIAO Teng(School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China;Key Laboratory of Urban Land Resources Monitoring and Simulation, MNR, Shenzhen 518000, China)

机构地区：[1]武汉大学测绘学院,湖北武汉430079 [2]自然资源部城市国土资源监测与仿真重点实验室,广东深圳518000

出　　处：《测绘学报》2020年第2期181-190,共10页Acta Geodaetica et Cartographica Sinica

基　　金：自然资源部城市国土资源监测与仿真重点实验室开放基金(KF-2018-03-025)~~

摘　　要：针对多视密集匹配的效率较低的问题,提出了GPU-PatchMatch多视密集匹配算法。该算法使用GPU提高PatchMatch的计算效率;同时充分利用稀疏场景信息,对深度信息进行规则初始化;为提高传播效率,使用了金字塔红黑板并行传播深度信息。最后在DTU、Strecha和Vaihigen数据集上进行了试验,并与常用的多视密集匹配算法进行对比。试验结果表明,本文算法在重建效率上有较大提高,与CPU算法(PMVS、MVE、OpenMVS)相比有7倍以上提升,与GPU算法相比也有2.5倍以上提升,表明本文算法的有效性。Aiming at the problem of low efficiency of multi-view dense matching,a GPU-PatchMatch multi-view dense matching algorithm is proposed.The algorithm uses GPU to improve the computational efficiency of PatchMatch.At the same time,it also makes full use of sparse scene information to initialize the depth information.In addition,in order to improve the propagation efficiency,it uses the pyramid red-blackboard to propagate the depth information in parallel.Finally,the experiments are carried out on the DTU,Strecha and Vaihigen datasets,and compared with the commonly used multi-view dense matching algorithms.The results show that our algorithm has a significant improvement in reconstruction efficiency,which is 7 times higher than the CPU algorithm(such as PMVS,MVE and OpenMVS),and 2.5 times higher than the GPU algorithm(such as Gipuma),which proves the effectiveness of the proposed method.

关 键 词：三维重建 多视密集匹 块匹配 并行计算 

分 类 号：P237[天文地球—摄影测量与遥感]