EPOS-Lid: Lidar benchmark dataset for pose estimation during non-cooperative rendezvous

Abstract

Lidar sensors are increasingly studied for space rendezvous applications, in particular for on-orbit servicing or active debris removal missions. With their active measurement principle, they provide accurate 3D point clouds which enable precise pose estimation. While simulated lidar data is relatively simple to generate under ideal conditions, real lidar point clouds can present high levels of noise and reflections. There is the need for representative lidar data to train and test pose estimation methods for non-cooperative space rendezvous scenarios. This work introduces EPOS-Lid, an openly available lidar benchmark dataset for this task. It comprises a synthetic dataset for training pose estimation methods, and real lidar point clouds collected at the European Proximity Operations Simulator (EPOS). Further, it is demonstrated with evaluation of benchmark methods how the datasets can be used for training and testing pose initialization and pose tracking methods.