Track a drone's position indoors where GPS is unavailable by fusing camera footage with IMU motion sensor data.
Benchmark a visual-inertial odometry system against standard research datasets like EuRoC or KITTI before deploying on custom hardware.
Enable GPS-correction on top of camera-IMU tracking to reduce long-distance drift in autonomous vehicle applications.
Build a robotics project that estimates camera position from live video without external positioning infrastructure.
Requires a matching Ubuntu and ROS version plus Ceres Solver, calibrated hardware with global-shutter cameras and hardware time-sync is needed for real-world use.
VINS-Fusion is a research system from Hong Kong University of Science and Technology that helps robots and vehicles figure out where they are in space. It does this by combining data from cameras and motion sensors to track position accurately, even in places where GPS is unavailable or unreliable. The technique is called visual-inertial odometry: the software watches the visual world through one or two cameras and simultaneously reads a motion sensor to estimate how the device is moving. By fusing those two data streams, it can maintain a precise location estimate in real time. The system supports several sensor combinations: a single camera paired with a motion sensor, two cameras paired with a motion sensor, or two cameras on their own. It can also pull in GPS data when available to correct for drift that accumulates over time with camera-only tracking. There is an optional loop closure feature, which means the system can recognize places it has seen before and correct its position estimate when it does, reducing errors that build up over long distances. VINS-Fusion was built for serious robotics and autonomous vehicle research. It runs on Ubuntu using ROS, which is a robotics software framework commonly used in academic and industrial robot development. Setting it up requires a matching version of Ubuntu, a working ROS installation, and a separate math library called Ceres Solver. The build process involves standard ROS tooling. A Docker container is also available for researchers who want to skip the manual dependency installation. The README includes detailed examples using two well-known research datasets: EuRoC, which contains drone flight recordings, and KITTI, which contains autonomous driving data. These examples show how to run the system with different sensor configurations and how to visualize the resulting path estimates. The project also notes that getting good results with your own hardware requires quality camera equipment, specifically cameras with global shutters and hardware time synchronization between sensors. This project is geared toward robotics researchers, PhD students, and engineers working on autonomous systems who need a working, tested implementation of visual-inertial state estimation to build on or benchmark against.
← hkust-aerial-robotics on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.