Image moment extraction based aerial photo selection for UAV high-precision geolocation without GPS

Abstract

Since GPS is susceptible to environmental interference or cannot be used in some specific situations, it is important to study the UAV geolocation under GPS-free condition. How to ensure that the aerial photos used for positioning have high accuracy is a pressing concern. The conventional approach relies on the number of matching feature points between aerial and satellite photos and verifying whether the aerial photo center projection falls within a designated area of the satellite photo. However, this method often yields suboptimal accuracy. This paper introduces a novel method for selecting aerial photos for geolocation based on an image moment extraction module. The image moment extraction module contains two parts: the relative error of the image invariant moment and the area relative error. The former is used to select the aerial photos with good performance in translation, rotation and scaling before and after projection, and the latter is adopted to reject aerial photos with excessive scaling before and after projection. Through experiments, we demonstrate the effectiveness and superiority of this approach. Results show that the image moment extraction module, when used for positioning with aerial photo center projection, significantly reduces the mean absolute error in UAV geolocation from 63.220 m to 9.454 m. The maximum error is reduced from 486.7 m to 47.194 m, with 70.96% of aerial photos suitable for positioning. This research is pivotal for achieving high-precision UAV geolocation in GPS-free environments.