IMU Preintegration for Visual-Inertial Odometry Pose Estimation

In this paper, based on Inertial Measurement Unit (IMU) preintegration, a pose estimation algorithm for Visual-Inertial Odometry (VIO) system has been investigated. Firstly, we developed the kinematic equation of acceleration and angular velocity of IMU. To avoid the re-integration caused by the change of the system states, the IMU measurements are preintegrated on the Riemannian manifold of rotation. Furthermore, we studied the propagation characteristic of noise and derived the state transition equation of the noise. So the influence of noise on pose estimation can be dramatically reduced. Compared with the famous Okvis algorithm, experimental results on two public datasets of EuRoC show that the proposed IMU preintegration algorithm achieves good estimation median of translation error with 0.14 m, which is better than Okvis with 0.24 m.