Creating High Quality DEMs of Large Scale Fluvial Environments Using Structure-from-Motion

During the past decade, advances in survey and sensor technology have generated new opportunities to investigate the structure and dynamics of fluvial systems. Key geomatic technologies include the Global Positioning System (GPS), digital photogrammetry, LiDAR, and terrestrial laser scanning (TLS). The application of such has resulted in a profound increase in the dimensionality of topographic surveys - from cross-sections to distributed 3d point clouds and digital elevation models (DEMs). Each of these technologies have been used successfully to derive high quality DEMs of fluvial environments; however, they often require specialized and expensive equipment, such as a TLS or large format camera, bespoke platforms such as survey aircraft, and consequently make data acquisition prohibitively expensive or highly labour intensive, thus restricting the extent and frequency of surveys. Recently, advances in computer vision and image analysis have led to development of a novel photogrammetric approach that is fully automated and suitable for use with simple compact (non-metric) cameras. In this paper, we evaluate a new photogrammetric method, Structure-from-Motion (SfM), and demonstrate how this can be used to generate DEMs of comparable quality to airborne LiDAR, using consumer grade cameras at low costs. Using the SfM software PhotoScan (version 0.8.5), high quality DEMs were produced for a 1.6 km reach and a 3.3 km reach of the braided Ahuriri River, New Zealand. Photographs used for DEM creation were acquired from a helicopter flying at 600 m and 800 m above ground level using a consumer grade 10.1mega-pixel, non-metric digital camera, resulting in object space resolution imagery of 0.12 m and 0.16 m respectively. Point clouds for the two study reaches were generated using 147 and 224 photographs respectively, and were extracted automatically in an arbitrary coordinate system; RTK-GPS located ground control points (GCPs) were used to define a 3d non-linear transformation to convert the point clouds to the absolute NZTM coordinate system, with average errors of 0.06 m in the horizontal and 0.11 m in the vertical dimensions. The final point clouds extracted had typical point spacings of 0.25 m, well above the metric resolution of airborne LiDAR. To improve data handling, the final point cloud was decimated to point spacings of 0.5 m using a recently developed gridding procedure (Rychkov, Brasington, & Vericat, 2012), and finally converted into a DEM using a Delaunay constrained TIN in ArcGIS. Results reveal SfM's ability to produce high quality terrain products of large scale fluvial environments that can outperform LiDAR, and can potentially compare with TLS. PhotoScan offers a straightforward method to generate, transform, and export DEMs that requires little user knowledge of photogrammetric processes. Further, the affordability and reduced field work offer low budget researchers the ability to produce repeat surveys for in-depth temporal studies. Funding supported by the New Zealand Department of Conservation.