Accuracy analysis of terrain point cloud acquired by “structure from motion” using aerial photos

The need to acquire high-quality digital topographic data is evident throughout geoscience research. The use of these data elevates the research level of geosciences. Airborne and terrestrial light detection and ranging (LiDAR) are currently the most prevalent techniques for generating such data, but the high costs and complex post processing of these laser-based techniques restrict their availability. In the past few years, a new stereoscopic photogrammetry mapping method called Structure from Motion (SfM) has been applied in geoscience, in which the 3D digital topography is reconstructed using feature matching algorithms from overlapping photographs of multiple viewpoints. SfM only needs a series of overlapping images with no special requirements about the camera positions, orientations and lens parameters, making it possible to use images collected from an affordable SfM platform to rapidly generate high-quality 3D digital topography. This paper summarizes the basic principles and the SfM workflow, and shows that SfM is a low-cost, effective tool for geoscience applications compared to LiDAR. We use a series of digital aerial photos with ~70% overlap collected at one-thousand-meter height to produce a textured (color) SfM point cloud with point density of 25.5/m². Such a high density point cloud allows us to generate a DEM with grid size of 0. 2m. Compared with LiDAR point cloud, statistical analysis shows that 58.3% of LiDAR points deviate vertically from the closed SfM point by <0.1 m and 88.3% by <0.2 m. There is different SfM accuracy in different landforms. The SfM accuracy is higher in low dips and subdued landforms than in steep landforms. In consideration of relative vertical error of 0.12 m in LiDAR data, SfM has a higher measuring accuracy compared with LiDAR.