RGB-D Dataset for Pose Estimation of Small UAS

Abstract

In this work, we present a new synthetic dataset for pose estimation of small unmanned aircraft systems (UAS) from RGB-D images. The dataset covers the full range of realistic poses and offers color images, depth information, masks, and pose annotations. A total of four distinct environments provide challenging conditions, as we show in our benchmarking study with the pose estimation method SAM-6D.We demonstrate the critical need for an upstream object detection method and investigate the influence of visible propellers on the accuracy of segmentations and pose estimates. The dataset is made publicly available to facilitate future research in the field.