On accuracy of position estimation from aerial imagery captured by low-flying UAVs

Accuracy analysis, Error propagation, Monte Carlo sampling, Planar scene, Image distortion

The application of low-flying Unmanned Aerial Vehicles (UAVs) for traffic monitoring and surveillance requires an estimation of position measurement accuracy of monitored objects. In this work we aim to analyse and provide insight into the accuracy of position estimation of objects based on aerial imagery captured by low flying UAV for the purpose of traffic monitoring. The analysis is focused on data gathered by a low-cost action camera mounted on a multicopter UAV flying above a planar scene. We assume a simple, straightforward method of position estimation of object based on homography mapping between two 2D planes derived from the position of images of landmarks - objects with known real world position. We assume errors caused by inaccuracy of the following values: the landmarks' real world position, the landmarks' and target's real world elevation, and the captured image positions of both the landmarks and the target. Additionally, a geometric deformation of captured frame caused by imperfect camera lenses is considered. The paper further analyses the effect of varying magnitude of errors, camera position, incidence angle and both the landmarks' and the target's positions in the scene or captured frame, and compares the results with real world experiments. The results can be used to estimate the feasibility and applicability of certain solutions to object position estimation. The tool for the calculation of the accuracy of position estimation and locating the most suitable camera pose for a given setup is provided.