Abstract
Structure-from-motion (SfM) in combination with multi-view stereo (MVS) represent techniques, which allow efficient generation of the point cloud from close-range photogrammetry (CRP) images of forest ground. Recent software products for the generation of digital terrain models (DTM) includes a wide range of interpolation methods. Previous studies showed different errors in elevations of DTMs interpolated with different methods. This study aims to analyze differences between the elevations of DTMs derived from CRP point cloud using different methods of interpolation. Six methods of interpolation included in modular system OPALS were tested in the study. In addition to simple methods of interpolation such as Snap or Moving average, more complex methods were used for interpolation of the DTMs elevations. For each method, 5 DTMs with resolution ranging from 1 to 20 cm were generated. Elevations of the DTMs were compared with the elevations of Global Navigation Satellite System (GNSS) surveyed check points. RMSE of DTMs elevations ranges from 3.4 cm to 16.2 cm. Differences between the elevations of DTMs interpolated using different methods and resolution were further investigated using one-way analysis of variance (ANOVA). The ANOVA rejected the statistical significance of the differences. Additionally, the spatial distribution of errors was analyzed. The analysis indicates that the interpolation of the extreme DTM values can be expected at the edges of the DTM when using the CRP images captured from single passing through the study site.