Abstract

The processing process to create a Digital Terrain Model (DTM) from image data of DJI Phantom 4 RTK is presented in this paper. The position of the camera recorded during the flight mission is processed in conjunction with the GNSS measurement data at the stationary station in the post-processing kinematic technique to achieve a highly accurate captured flight position. The 3D image model is built from Overlapping 2D images using the Structure from motion (SFM) technique. Three highly accurate measured ground control points are used to accurately georeference the 3D image model. The overall alignment error of the 3D model is approximately 2.7 cm. From the 3D model, the dense ground points were determined based on the combination of Classify Points technique and Classify Ground Points technique and used to create the DTM model. Five feature points at unobstructed locations were used to evaluate positional accuracy having an overall position error of approximately 2.5 cm. Fourteen feature points at occluded locations were used only for elevation error estimation having a standard deviation error of approximately 0.147 m. The results show that i) at an unobstructed location, the images taken from the Phantom 4 RTK can enable the creation of 3D surfaces with centimetre-level accuracy, and ii) at the obscured location, the accuracy in DTM generation depends on the classification method that identifies points belong to terrain surface.