照片数量对SfM动床河工模型地形三维重构的影响研究

Effect of input number of photos on 3D digital elevation data reconstruction by structure from motion in river model test with a moveable bed

SfM是摄影测量领域近年来飞速发展的三维场景重构技术,目前已在地貌勘测、文物勘察、冰川演化、岩石勘探、土壤侵蚀等研究领域开展了大量尝试和应用,而在河工模型方面的应用研究尚处于起步阶段。本文开展了输入照片数量对SfM法动床河工模型地形三维数字高程地形(DEM)重构精度影响研究,采用4个不同视角对河工模型动床冲淤地形进行拍摄,选取不同数量照片作为输入开展SfM法三维地形DEM重构,然后将每组DEM重构结果与传统测针法测量进行对比。结果表明,当照片数量较少时,增加照片数量可改善SfM重构精度,但当照片数量较多时,虽可为SfM提供更多的床面变化图像信息,但同时带来由于特征点成像变化而引起的误匹配现象,反而可能导致SfM重构精度下降。鉴于不同动床模型的模型沙和光照等外部条件不同,实际应用时须针对不同模型的特点进行深入对比和分析,选取合理的照片张数开展SfM法动床冲淤地形测量。

SfM(Structure from Motion) is one of the rapidly developing three-dimensional structure reconstruction technologies in the field of Photogrammetry in recent years, which has been extensively verified and applied in geomorphology survey,cultural relics survey,glacier evolution,rock exploration,soil erosion,and other different research fields. In this study,the effect of the input number of photos on DEM(Digital Elevation Model) reconstruction by SfM for river model test with a mobile bed was studied experimentally. The photos were taken from 4 different views around the research area of a physical model. Different numbers of photos were selected as the input information to reconstruct DEM by using the SfM method.The DEM data of each run was compared with the elevation measured by the traditional point gauge method. The results show that when the numbers of photos are small,increasing photos may improve the SfM results, whereas when the numbers of photos are large, further increasing photos can provide SfM with more information about bed variation, but at the same time, it also brings the mismatching phenomenon caused by the change of feature points,which on the contrary leads to the decrease of SfM reconstruction accuracy. Due to the different conditions such as types of sediment,illumination condition,and so on,it is necessary to analyze the characteristics of the specific model in depth to select a reasonable number of photos for carrying out the SfM topographic survey.