Measuring 3D-reconstruction quality in probabilistic volumetric maps with the Wasserstein Distance

Conference: ISR Europe 2023 - 56th International Symposium on Robotics
09/26/2023 - 09/27/2023 at Stuttgart, Germany

Proceedings: ISR Europe 2023

Pages: 7Language: englishTyp: PDF

Authors:
Aravecchia, Stephanie; Pradalier, Cedric (IRL2958 GT-CNRS, Metz, France)
Richard, Antoine (University of Luxembourg, Luxembourg)
Clausel, Marianne (Universit´e de Lorraine, Nancy, France)

Abstract:
In this study, we address the challenge of measuring 3D-reconstruction quality in large unstructured environments, when the map is built with uncertainty in the robot localization. The challenge lies in measuring the quality of a reconstruction against the ground-truth when the data is extremely sparse and where traditional methods, such as surface distance metrics, fail. We propose a complete methodology to measure the quality of the reconstruction, at a local level, in both structured and unstructured environments. Building upon the fact that a common map representation in robotics is the probabilistic volumetric map, we propose, along this methodology, to use a novel metric to measure the map quality based directly on the voxels’ occupancy likelihood: the Wasserstein Distance. Finally, we evaluate this Wasserstein Distance metric in simulation, under different level of noise in the robot localization, and in a real world experiment, demonstrating the robustness of our method.