Modern unmanned aerial vehicles rely on the global positioning system to carry out autonomous missions and return to launch if the control link is lost. GPS reception can be jammed or lost due to several natural, intentional, and electromagnetic causes, which disables autonomous systems.
The purpose of this paper was to research localization options of unmanned aerial vehicles in those scenarios. Based on the theory of autonomous mobile systems, computer vision, and embedded systems, we proposed five localization methods without GPS reception: inertial odometry, odometry with quadcopter model, two approaches with visual odometry, and using SLAM. We have tested the performance of all five methods on two datasets, that we obtained using a test quadcopter platform and data captured by its sensors. We have used multiple metrics to evaluate the accuracy, precision, and usability of the proposed methods.
We have concluded that all five methods are suitable for real-life use in the case of GPS signal absence, except for inertial odometry, which is not accurate enough. Optical odometry methods performed the best, but they require good ground-camera distance measurement. Odometry with quadcopter model worked well but does not give accurate estimates in windy environments. All methods except SLAM are appropriate for use on embedded systems with limited resources.
|
