A real-time visual odometry pipeline using stereo event-based cameras, leveraging Time Surfaces for direct geometric mapping and tracking.
ESVO is an open-source visual odometry system that estimates camera motion and reconstructs 3D environments using stereo event-based cameras. It processes asynchronous event streams through a Time Surface representation to provide real-time, low-latency pose estimation without relying on traditional frame-based images. The system solves the problem of visual odometry in high-speed, dynamic scenarios where conventional cameras struggle.
Robotics researchers and engineers working with event-based cameras, particularly those developing autonomous systems, drones, or mobile robots that require real-time visual odometry in challenging lighting and motion conditions.
Developers choose ESVO for its direct geometric approach to event-based visual odometry, offering real-time performance with stereo event cameras. Its unique selling point is the unified Time Surface representation that enables efficient mapping and tracking from raw event data, making it particularly suitable for high-speed applications where traditional frame-based methods fail.
This repository maintains the implementation of "Event-based Stereo Visual Odometry".
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Achieves real-time odometry on modern multi-core CPUs, with optimizations allowing up to 200 fps on embedded systems like Jetson TX2, as noted in the project logs.
Uses a geometric approach with Time Surfaces to process raw event data directly, avoiding frame-based intermediates and enabling low-latency motion estimation for dynamic scenes.
Leverages stereo camera geometry for accurate 3D reconstruction and depth estimation, supported by a dedicated multi-view stereo module for mapping with known poses.
Based on peer-reviewed publications from top venues like IEEE T-RO and ECCV, ensuring robust and validated methods for event-based visual odometry.
Requires specific stereo event cameras (e.g., DAVIS) and installation of proprietary drivers from rpg_dvs_ros, along with a full ROS stack, making setup time-consuming and error-prone.
Heavily integrated with ROS for data handling and execution, necessitating ROS expertise and adding infrastructure overhead that may not suit lightweight or non-ROS applications.
Results vary between runs due to stochastic operations and parallel processing, as admitted in the README, which complicates reproducible testing and deployment.