Accuracy Assesment of direct sensor orientation in UAV Photogrammetry

Abstract

With the advent of Unmanned Aerial Vehicles, it becomes possible to use these vehicles to acquire photogrammetric images in a more cheap and flexible way compared with conventional photogrammetry from a full-sized airplane. A big factor in costs and resources in conventional photogrammetry is signalling and measuring ground control points, which are necessary to determine the location and attitude of the photos. Eliminating the need for ground control points would make the photogrammetric process much less costly and labour intensive. A UAV contains navigational sensors (a GPS receiver and a Inertial Measurement Unit) for it's own navigation. The output of these sensors can also be used for the orientation of the photographs. However, the accuracy of these sensors, especially low cost off-the-shelf sensors, is limited. This thesis concentrates on determining the accuracy of low-cost navigational sensors and the effect of inaccuracies in these sensors on the result of the photogrammetric product: a Digital Terrain Model. Using a UAV developed by the company Heering UAS, founded by Pieter Wijkstra, the effects of inaccuracies in the GPS receiver and the IMU in the UAV on the outcoming Digital Elevation model has been tested. First the theoretical effects of inaccuracies of navigation sensors on the final product is investigated, using the photogrammetric collinearity equations. To validate the theoretical results, a number of tests has been carried out using an available platform with an off-the-shelf camera and dito navigation sensors. A first test, using flight data and photos taken at the unfinished A4 highway site just south of Delft, shows the sensitivity of the final model to errors in the navigational sensors. This test did not give enough data though to give enough insight in the effects of navigational sensor errors. Therefore a second test was carried out using a larger photoblock flown at the dyke of the river Lek at Ammerstol. This test showed that the errors in the terrain model caused by errors in the on-board GPS receiver follow the results of the theoretical simulation. However, because at the time, the UAV did not contain a reliable IMU, the effect of IMU errors could not be tested. So a third test was conducted, using an off-the-shelf IMU combined with a high accuracy GPS receiver to asses the effect of errors in the IMU on the resulting terrain model. The results show that off-the-shelf navigation sensors are not accurate enough to produce a reliable end result. Therefore eliminating ground control points would result in a final product which is not accurate enough. Using the available navigation sensors it is possible to reduce the amount of ground control points.