Visual-Inertial Odometry

Fusing a camera and an IMU to continuously estimate a device’s own motion trajectory.

Visual-inertial odometry (VIO) fuses camera images with IMU (inertial measurement unit) readings to continuously estimate a device’s own position and orientation; using only a camera is called visual odometry (VO). The two sensors complement each other: the IMU samples fast and keeps working even during violent motion, but drifts quickly once its readings are integrated; the camera constrains that drift, but is thrown off by motion blur and texture-less surfaces, and a monocular camera alone doesn’t know the true scale — the gravity and acceleration the IMU measures can restore that scale. Implementations split into filtering-based and optimization-based approaches (such as VINS-Mono), and separately into loosely-coupled and tightly-coupled fusion. VIO only performs local estimation, so error still accumulates over a long run; adding loop closure detection and global optimization turns it into visual-inertial SLAM. It is commonly used to localize drones, AR devices, and legged robots in environments without GPS.

ExampleA drone flying inside an indoor warehouse has no GPS signal, so it relies on its onboard camera and IMU running VIO to know where it has flown in real time.

Also called

VIO, Visual-Inertial Navigation

Related

Visual Odometry · Inertial Measurement Unit · VINS-Mono / VINS-Fusion · IMU Preintegration · Tightly-Coupled vs. Loosely-Coupled Fusion · Visual SLAM

Sources

Visual odometry (Wikipedia)
VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator (arXiv)