Relative Visual Localization for Unmanned Aerial Systems

TL;DR

This paper introduces a novel visual relative localization system for UAVs using a spherical camera and fiducial markers, enabling accurate pose estimation in GPS-denied environments with efficient computation.

Contribution

The paper presents a new spherical camera-based localization method with an algorithm to optimize image processing for real-time UAV tracking.

Findings

Accurate UAV tracking in GPS-denied environments demonstrated

System achieves real-time performance with optimized image selection

Validation against motion capture confirms high localization accuracy

Abstract

Cooperative Unmanned Aerial Systems (UASs) in GPS-denied environments demand an accurate pose-localization system to ensure efficient operation. In this paper we present a novel visual relative localization system capable of monitoring a 360$^o$ Field-of-View (FoV) in the immediate surroundings of the UAS using a spherical camera. Collaborating UASs carry a set of fiducial markers which are detected by the camera-system. The spherical image is partitioned and rectified into a set of square images. An algorithm is proposed to select the number of images that balances the computational load while maintaining a minimum tracking-accuracy level. The developed system tracks UASs in the vicinity of the spherical camera and experimental studies using two UASs are offered to validate the performance of the relative visual localization against that of a motion capture system.