Online Time Calibration method for Monocular Visual-Inertial Odometry based on Improved Adaptive Extended Kalman Filter

This paper focuses on the time delay estimation of the Visual-Inertial Odometry (VIO) system. VIO has been widely used in VR/AR, unmanned driving and mobile robots due to its low cost, small size and abundant information. However, the time delay between the IMU and the camera, such as trigger delay and transmission delay, and the lack of accurate synchronization clock affects the performance of VIO system. Therefore, the time delay estimation plays a key role in improving system performance. The traditional extended Kalman filter (EKF) is not accurate because of the unknown noise. The adaptive extended Kalman filter (AEKF) can estimate and correct the statistical characteristics of noise. However, AEKF is easy to fall into local optimum and converges slowly due to excessive reliance on historical data. This paper proposes an improved adaptive extended Kalman filter method, which adds a forgetting factor on the basis of AEKF.It increases the weight of new observations by increasing the one-step prediction covariance matrix P. The results show that the proposed method can enhance the estimation speed, effectively improve the positioning accuracy and adaptability of the VIO system.