Spatiotemporal Calibration Based on Nonlinear Optimization for Heterogeneous Information Including GNSS Raw Data

Abstract

Achieving high-precision positioning through multi-source integration has become an inevitable trend in autonomous vehicle systems, and the spatiotemporal calibration of multi-source information is the primary prerequisite. This paper proposes a spatiotemporal calibration algorithm for the fusion system of GNSS data, LiDAR data, and visual data with the inertial sensor as the central coordinate system. First, we use the pseudo-distance information of GNSS to construct the space-time calibration model of GNSS (Global Navigation Satellite System) relative to IMU (Inertial Measurement Unit). Second, based on the reprojection principle, we construct a spatiotemporal calibration model of visual images relative to the IMU. Then, according to the distance formula of the LiDAR (Light Detection and Ranging) points cloud, the space-time calibration model of the LiDAR points cloud relative to the IMU is established. Finally, we use the nonlinear optimization algorithm to obtain the spatiotemporal parameters. We have done extensive simulations based on simulated data and publicly available real-world datasets. The simulation results show that using the proposed calibration model yields spatiotemporal parameter accuracy superior to existing calibration algorithms and exhibits some degree of robustness to the noise in IMU data. It achieves approximately 40% improvement in position estimation accuracy with the open-source odometry and the real-world datasets while ensuring good safety and reliability under high computational efficiency.