Vehicle-Motion-Constraint-Based Visual-Inertial-Odometer Fusion With Online Extrinsic Calibration

Autonomous positioning in complex urban areas is a challenging problem and has attracted increasing attention in academia. Multisensor fusion of camera, inertial measurement unit (IMU), and wheel odometer has been a prevailing solution that is not only low-cost but easy-to-build. Most of the related methods fuse the odometer measurements into the visual-inertial framework by preintegrating the wheel speed into the displacement. However, they are not very suitable for some scenarios where the vehicle velocity changes frequently. Instead of preintegrating the odometer measurements, in this article, we present an efficient and practical visual-inertial-odometer approach that fuses the wheel speed into the visual-inertial framework directly based on the vehicle motion constraint. Specifically, we use the nonholonomic constraint (NHC) and lever-arm compensation to introduce the vehicle velocity measurement into the fusion framework to limit the drift. Meanwhile, the proposed framework also allows to online calibration of the IMU-odometer extrinsic parameters (EPs) explicitly. Moreover, we develop an observability-aware method to enhance the system stability and the performance of online extrinsic calibration. We also develop a robust initialization method to obtain the initial values of the visual-inertial-odometer system in one-step optimization. Our approach is validated extensively in simulation environments, autonomous driving public datasets, and real-world experiments. The simulation results show the extrinsic calibration error is within 0.1°/0.02 m for rotation and translation. The public dataset and real-world ground robot experiments show a 0.3% position error in both the 4-km long urban area route and the 403-m long park route.