Application of Visual Inertial Odometry for Pose Estimation of a Mobile Robot

In this paper, an application of visual inertial odometry is described. Details of the implementation including the calibration are also described. The method uses visual data from two cameras and the measurements from inertial measurement unit. Previous researches usually used only front view image, whereas this study uses three different visual data from different view angles: front view, floor view, and ceiling view. The results are compared and the effect according to different view angles is analyzed. The comparison shows that the number of detectable features affects the performance of the visual odometry.