A comprehensive uncertainty analysis and method of geometric calibration for a circular scanning airborne lidar

Michael Oliver Gonsalves

Abstract

This dissertation describes an automated technique for ascertaining the values of the geometric calibration parameters of an airborne lidar. A least squares approach is employed that adjusts the point cloud to a single planar surface which could be either a narrow airport runway or a dynamic sea surface. Going beyond the customary three boresight angles, the proposed adjustment can determine up to eleven calibration parameters to a precision that renders a negligible contribution to the point cloud's positional uncertainty. Presently under development is the Coastal Zone Mapping and Imaging Lidar (CZMIL), which, unlike most contemporary systems that use oscillating mirrors to reflect the beam, will use a circular spinning prism to refract the laser in the desired direction. This departure from the traditional scanner presents the potential for internal geometric misalignments not previously experienced. Rather than relying on past calibration practices (like requiring data be acquired over a pitched-roof), a more robust method of calibration is established which does not depend on the presence of any cultural features. To develop this new method of calibration, the laser point positioning equation for this lidar was developed first. The system was then simulated in the MATLAB environment. Using these artificial datasets, the behavior of each geometric parameter iii was systematically manipulated, understood and calibrated, while an optimal flight strategy for the calibration acquisition was simultaneously developed. Finally, the total propagated uncertainty (TPU) of the point cloud was determined using a propagation of variances. Using this TPU module, the strength of the calibration solution was assessed. For example, four flight lines each of 20 seconds in duration contained sufficient information to determine the calibration parameters to such a degree of confidence that their contribution to the final point cloud uncertainty was only 0.012m in the horizontal and 0.002m in the vertical (1σ).