An event-based stereo 3D mapping and tracking pipeline for autonomous vehicles - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

An event-based stereo 3D mapping and tracking pipeline for autonomous vehicles

Résumé

Event cameras are bio-inspired, motion-activated sensors which generate asynchronous events instead of intensity images at a fixed rate. These sensors have been shown to outperform traditional frame-based cameras by large margins, in case of high-speed motions and scenes with high dynamic range. Next-generation intelligent vehicles are expected to greatly benefit from these novel cameras, especially in adverse lighting conditions, and their potential is still largely untapped. In the last decade, the continuous stream of events produced by an event camera has been exploited in numerous 3D perception tasks (depth estimation, 6-DoF tracking, visual-inertial odometry, etc.). In this paper, we propose an event-based stereo pipeline for simultaneous 3D mapping and tracking. The mapping module relies on DSI (Disparity Space Image) fusion, and the tracking module makes use of time surfaces as anisotropic distance fields, to estimate the pose of the stereo camera. Numerical experiments with a publicly-available event dataset recorded by a car in different urban environments, show the effectiveness of the proposed architecture.
Fichier non déposé

Dates et versions

hal-04323607 , version 1 (05-12-2023)

Identifiants

  • HAL Id : hal-04323607 , version 1

Citer

Anass El Moudni, Fabio Morbidi, Sebastien Kramm, Rémi Boutteau. An event-based stereo 3D mapping and tracking pipeline for autonomous vehicles. Journée inter GdR ISIS et Robotique "Caméra à événements appliquée à la robotique", Nov 2023, Paris, France. ⟨hal-04323607⟩
22 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More