Mapping Accuracy of Close-Range Terrestrial Digital Photogrammetry

Abstract

Recently, close-range terrestrial digital photogrammetry has received increased attention in geomorphological studies due to high image resolution, sufficient accuracy, and cost efficiency compared to the other techniques. Digital photogrammetry can be described as a non-contacting remote sensing technique that facilitates earth surface reconstruction and provides digital coordinates of the earth points and geospatial features based on a series of overlapping images. It uses Structure-from-Motion and Multi-view Stereo algorithms to recover 3-D features from a projected 2-D scene of the collected images. The final products are 3-D point cloud and digital surface model (DSM). This study focused on evaluating the accuracy of the developed DSM for an artificial channel in a 4 m x 1 m soil bed manufactured in a controlled lab environment (Figure 1). <fig><graphic xlink:href=23028_files/23028-08.jpg id=ID_187e99a1-64a5-4fdb-a18f-c6bdae1eabbd></graphic></fig> The factors that were evaluated were number and spatial distribution of ground control points (GCP). It was found that in all experiments the errors at the control points can be as high as 3.1 cm in the X and Y directions and up to 7 cm in the Z direction. The accuracy improved up to the best average errors of 1.2 cm (X, Y) and 2.1 cm (Z) for the case of random GCP placement and optimal density of ~2.5 GCP per image. The accuracy decreased with either lower number of GCPs or their biased placement along either side of the channel. Error and group analysis showed that the accuracy of the photogrammetric DSM was affected by biased GCP placement, especially when clustered toward one area. Higher resolution of the point cloud may become important for more accurate identification of rapid elevation changes in a DSM as well as channel overall geometry. Although the method and produced DSMs can be used for the estimation of the elevation changes in field-based soil erosion studies, the generated errors (especially in the vertical dimension) may be comparable to the elevation change itself. Thus, this technique must be used with caution when the expected elevation change is small.