Improved Preintegration Method for GNSS/IMU/In-Vehicle Sensors Navigation using Graph Optimization

GNSS/IMU/in-vehicle sensors navigation can provide accurate localization for land vehicles. Although the sensor fusion is often carried out by Kalman filter, graph optimization can obtain better state estimation. However, the graph optimization is still affected by the lack of rigorous IMU/in-vehicle sensors preintegration. In this paper, an improved preintegration method for GNSS/IMU/in-vehicle sensors navigation using graph optimization is proposed. Different from current methods, the preintegration that involves total error parameters of in-vehicle sensors is derived. The improved factors are subsequently fused with GNSS measurements and kinematic models using graph optimization, which can allow to yield a total parameters estimation. Thereby, more accurate relative constraints provided by the proposed method can improve the positioning accuracy compared with traditional preintegration methods. Meanwhile, the estimation accuracy of the proposed method for these error parameters can be superior to Kalman filter. The KITTI datasets and field tests are carried out to validate the competitive performance of the proposed method. Finally, the computational load of the proposed method is also discussed to verify its applicability for real-time operations.