Temporal delay estimation of sparse direct visual inertial odometry for mobile robots

Abstract

Most visual inertial odometry(VIO) system can get an ideal result on datasets, which typically use an external device to synchronize measurements between the camera and the inertial measurement unit(IMU). But for an inexpensive homemade Cameras-IMU system, the accuracy of these algorithms usually degrades or even fails because of the temporal misalignment between the camera and the IMU. In this article, we focus on the time synchronization problem in direct VIO system, and we propose a method to calibrate the temporal delay between the camera and the IMU. To achieve this, the temporal delay parameter is added into the state variable of the extended Kalman filter(EKF), and the brightness error of the landmark is used as the error function to update the EKF filter. Finally, we construct a sparse direct VIO system with online temporal delay estimation, and make a comparison with the VINS-mono and the ROVIO. The experiments show that the brightness error of the landmark can be used to accurately calibrate the temporal delay between the image and the IMU measurements in direct VIO system, and the method considering the temporal delay will improve the performance of the direct VIO system.