基于无人机序列图像的多视图几何三维重建研究

Multiview Geometric 3D Reconstruction Based on UAV Sequence Images

基于无人机序列图像的多视图几何三维重建主要包括特征提取与匹配、稀疏和密集点云重建、纹理映射等几个步骤。重建的方式主要有非增加式和增加式,其中,增加式三维重建对最初始视图重建的精度要求非常高,并且在视图数量不断增加的时候,要不断地进行捆集调整,因此效率不好且精度无法得到保证。非增加式三维重建则通过全局方式计算目标三维点云,不需要多次进行捆集调整,且对初始视图重建的精度要求不高,在效率和精度方面优于增加式,因而成为研究热点。在OpenCV、OpenMVG等开源方式下,编程实现了非增加式的多视图几何三维重建,并用3组不同的无人机影像进行实验,实验结果表明该方式重建速度快、精度高,具有很强的实用性。

Multiview geometric 3D reconstruction based on UAV sequence images includes severalsteps： feature extraction and matching, sparse and dense point cloud reconstruction, texture mappingand so on. Reconstruction methods are mainly nonincremental and incremental, among them, incremental is gradually gain view, since it needs constant cluster adjustment in the meantime, it is inefficient and inaccurate. nonincremental is globally computing 3D cloud point and only global clusteradjustment. it is superior to incremental both on time and efficiency. Therefore, the 3D reconstructionof non - incremental become a hot research. Programming achieve the system of the 3D reconstruction of nonincremental with the open source of OpenCV, OpenMVG. Three groups of different UAVimages were used to carry out experiments. The experimental results show that this method is fast, accurate and practical.