[G05-P] Poster

[G05-P-02] Evaluation of digital surface models created from LiDAR and optical sensor data collected with unmanned aerial systems

^○Andrzej Borkowski, Grzegorz Jozkow, Agata Walicka, Mateusz Karpina, Przemyslaw Tymkow (Wroclaw University of Environmental and Life Sciences, Wroclaw, Poland)

Unmanned aerial systems (UASs) are becoming more popular in local scale topographic mapping. The state of the art in topographic UAS mapping is the use of consumer-grade RGB cameras, however, sensor development caused that laser scanners started to be used for this purpose. This work aims on comparative assessment of digital surface models (DSMs) created form data collected with 3 various UAS sensors: Nikon D800 RGB camera, Optris PI Lightweight 450 thermal camera, and Velodyne HDL-32E laser scanner. The evaluation focused on assessing the impact of georeferencing method to created DSM. All data was processed using standard methods. In the case of imagery data a bundle block adjustment (BA) was executed and was followed by image dense matching to create point cloud necessary for DSM modeling. Three types of georeferencing constraints in BA were investigated: ground control point (GCPs) measured with GNSS-RTK technique, image projection centers obtained from onboard GNSS-RTK measurements, and from onboard navigational GNSS solution. In the case LiDAR data, direct georeferencing of the point cloud using post-processed kinematic GNSS/INS solution was executed. Obtained DSMs were compared to the most accurate DSM created from high resolution RGB images georeferenced using GCPs. Results showed that georeferencing method has higher impact to the DSM accuracy than the quality of the sensor. Low quality thermal images georefer-enced using GCPs resulted in DSM of vertical RMSE equal to 17 cm, while good quality RGB images georeferenced with other methods resulted in vertical RMSE equal to 60 cm, however, a significant bias of about 50 cm was observed. It proved that internal accuracy of the point cloud created from RGB images is very high. In the case of LiDAR data, obtained RMSE was equal to 47 cm, but the internal accuracy of the point cloud was much lower than for RGB data. It was caused by lower quality of GNSS/INS solution, especially the orientation.