Online Visual-Inertial Extrinsic Calibration Utilizing GNSS Measurements for Vehicle Applications

Abstract

The visual-inertial navigation system (VINS) is a low-cost sensor suite that utilizes inertial measurement unit (IMU) and camera data for vehicle ego-motion estimation and environmental perception. For VINS, the accurate calibration of the inter-sensor (IMU-camera, camera-camera) extrinsic parameters is a prerequisite for optimal information fusion. Currently, modern vehicles are usually equipped with global navigation satellite system (GNSS) devices, which provide driftless information of the vehicle position and could be utilized to enhance inter-sensor extrinsic calibration. In this paper, we propose an online method for visual-inertial extrinsic calibration by fusing IMU data, visual observations and GNSS raw measurements (i.e., pseudorange and carrier phase measurements) in a tightly coupled way. To evaluate the proposed method, both simulation tests and real-world experiments under typical vehicular scenarios were conducted. The results verified the GNSS aiding effect on visual-inertial extrinsic calibration, especially for the IMU-camera relative rotation and the inter-camera baseline. Based on the 60-second calibration results, the visual-inertial odometry (VIO) showed superior relative positioning performance, achieving 0.87 m height drift and 0.331% median short-term translation error over a 2380 m trajectory, which is in contrast 3.81 m and 0.785% without GNSS aiding during calibration. The proposed method provides an efficient and low-cost way to calibrate the inter-sensor extrinsic parameters via making full use of the vehicle-mounted sensors, which could be used for either factory calibration or real-time multi-sensor fusion applications.