After visual pre-selection, approximate coordinates were assigned to the photographs based on the synchronised GPS flight path. Geotagging4 software was used to write the coordinates to image JPEG EXIF headers. Agisoft Photoscan Professional (0.85) software was used for the 3D reconstruction of the camera positions and terrain features. The specific algorithms implemented in Photoscan are not detailed in the manual, however, a description of the SfM procedure in Photoscan and commonly used parameters are described in Verhoeven (2011). Photoscan follows a common SfM and multi view stereopsis (MVS) workflow starting with image feature identification and feature matching. The approximate GPS coordinates of the camera stations were used at this stage to guide the matching process. Image matching was carried out with the PhotoScan accuracy set to high. The initial bundle adjustment resulted in a position and orientation for each camera exposure station and the 3D coordinates of all image features. These coordinates formed a sparse 3D point cloud of the terrain (1.85 million points). A dense geometry reconstruction based on multi-view stereopsis resulted in a more detailed 3D model with 10 million facets. This model was used to identify the 12 large GCPs to improve the absolute accuracy of the bundle adjustment and the 3D model. After the recomputed bundle adjustment a new 3D model with50 million facets was generated with the PhotoScan accuracy for the geometry build set to high. Finally, a DSM was generated by gridding the 3D model based on a given cartographic projection(WGS84 UTM 49S) and cell size (2 cm). Projecting the original photographs onto the 3D surface and blending their overlap zones produced an orthophoto mosaic (orthomosaic) of the whole area.
Due to the relatively low flying height we were able to generate a DSM at 2 cm resolution and an orthomosaic at 1 cm. An assessment of the geometric accuracy in easting (X), northing(Y), and height (Z) was carried out for the orthomosaic and DSM. Thirty of the small orange GCPs (excluded from georeferencing in the SfM process) were identified in the orthomosaic. The coordinates of the disk centroids were retrieved from the image mosaic and compared to the corresponding surveyed GPS coordinates, resulting in mean and root mean squared error (RMSE) accuracy measures in the X and Y direction. The height value was derived from the DSM for the GCP centroids and also compared to the GPS observations, producing mean and RMSE accuracy measures for the Z direction ( Höhle and Höhle , 2009; Harwin and Lucieer, 2012).