Performance Analysis of Visual-Inertial Navigation System with Feature Track Parameters

In this paper, we study the estimation performance of a visual-inertial navigation system (VINS) with two visual feature parameters: the number of visual feature tracks and vertical resolution of an image. While VINS have been intensively researched in the recent two decades, how quantitatively these parameters affect the estimation performance is still unclear. We try to answer this question by designing numerical simulation that is based on the real-world sensor setup. Specifically, we design two simulation experiments with decreasing number of visual features to simulate a visually challenging environment and decreasing vertical resolution of an image to mimic a restricted field of view of a ground vehicle with a frontal view. Our simulation results reveal that the 90% of visual feature loss induces 4.37 and 2.03 error gain for the position and attitude, respectively. Furthermore, the study shows that the vertical resolution has significant negative effect on the vertical position estimates.