Dense RGB-D mapping of large scale environments for real-time localisation and autonomous navigation

Maxime Meilland (1) , Patrick Rives (1) , Andrew I. Comport (2)

1 AROLAG - Advanced Image Understanding and Autonomous Systems
2 Laboratoire d'Informatique, Signaux, et Systèmes de Sophia-Antipolis (I3S) / Equipe SYSTEMES

This paper presents a method and apparatus for building 3D dense visual maps of large scale environments for real-time localisation and autonomous navigation. We propose a spherical ego-centric representation of the environment which is able to reproduce photo-realistic omnidirectional views of captured environments. This representation is composed of a graph of locally accurate augmented spherical panoramas that allows to generate varying viewpoints through novel view synthesis. The spheres are related by a graph of 6 d.o.f. poses which are estimated through multi-view spherical registration. It is shown that this representation can be used to accurately localise a vehicle navigating within the spherical graph, using only a monocular camera for accurate localisation. To perform this task, an efficient direct image registration technique is employed. This approach directly exploits the advantages of the spherical representation by minimising a photometric error between a current image and a reference sphere. Autonomous navigation results are shown in challenging urban environ- ments, containing pedestrians and other vehicles.

Domaines: Robotique [cs.RO]

Maxime Meilland, Patrick Rives, Andrew I. Comport. Dense RGB-D mapping of large scale environments for real-time localisation and autonomous navigation. Intelligent Vehicle (IV'12) Workshop on Navigation, Perception, Accurate Positioning and Mapping for Intelligent Vehicles, Jun 2012, Alcala de Henares, Spain. ⟨hal-00752897⟩