Using fiducial markers to improve localization of a drone

Abstract

Localization is a widely known task in a world of robotics. Mobile robots nowadays are becoming fully autonomous. For a robot to be fully autonomous, the ability to localize itself is very fundamental. Sufficient accuracy is often achieved by using multiple sources of position information. In the scope of this article, two such sources were used: Intel RealSense T265 tracking camera for continuous visual odometry calculation and RGB sensor in combination with ArUco fiducial markers to minimize error accumulated in visual odometry. These markers - just like all sources of information - have their flaws. To minimize the error caused by flaws of ArUco markers, variance filter was successfully implemented and evaluated. The whole system was tested several times on a drone, which was able to fly autonomously in an indoor environment. Furthermore, all processes needed for such localization ran smoothly on the Nvidia Jetson Nano computer mounted directly on the drone while flying. All these achievements were proven to be a really strong foundation for an autonomous drone, which can later perform useful tasks in indoor environments, including warehouse inventories, industrial halls inspections, or even some security operations. All this without an operator or pilot intervention.