Camera-IMU Calibration

Finds the relative pose and time offset between a camera and an IMU so their data can be aligned and fused.

Visual-inertial odometry (VIO, which estimates its own motion using a camera plus an IMU) needs to process both sensors’ data together, which requires knowing their spatial extrinsics (the IMU’s rotation and translation relative to the camera) and their time offset (the fixed delay between the two devices’ timestamps caused by triggering and transmission); solving for these two things is camera-IMU calibration. The most widely used offline tool is ETH’s open-source Kalibr, based on work by Furgale and colleagues at IROS 2013: it represents the motion trajectory with a continuous-time B-spline, has the device waved thoroughly along every axis in front of an Aprilgrid calibration board, and jointly optimizes the extrinsics and time offset; the IMU’s noise parameters are usually measured beforehand with Allan variance and fed in as input. Systems like VINS-Mono can also estimate the time offset online while running (Qin and Shen, IROS 2018). When calibration is inaccurate, VIO drifts noticeably or can even diverge.

ExampleTo calibrate the RealSense D435i on a handheld data-collection rig: record a data bag of translating and rotating in front of an Aprilgrid, use Kalibr to solve for the camera-to-IMU transform matrix and the time offset (usually on the order of milliseconds), and write it into VINS-Fusion’s configuration file.

Also called

Visual-Inertial Calibration, Camera-IMU Extrinsic and Time-Offset Calibration

Related

Visual-Inertial Odometry · Kalibr · Inertial Measurement Unit · Camera Extrinsics · Multi-sensor Time Synchronization / Timestamp Alignment · Allan Variance

Sources

ethz-asl/kalibr（GitHub）
Online Temporal Calibration for Monocular Visual-Inertial Systems (arXiv 1808.00692, IROS 2018)