Stable Estimation of Vehicle Ego-motion for Monocular Camera Sensors

Abstract

In this paper, we propose a stable vehicle ego-motion estimation method for monocular camera sensors. Vehicle ego-motion is the motion of a camera fixed to a vehicle. The 8-point-algorithm is well known as the typical method for obtaining camera motion, however, it outputs unreliable results in some cases. To obtain stable results, several methods have been proposed using a road surface constraint, however, there are difficult scenarios for these methods such as non-flat or non-textured roads. We solved these problems by using a constraint derived from vehicle motion.