Visual odometry based on the Fourier-Mellin transform for a rover using a monocular ground-facing camera

Abstract

This paper presents a visual odometry method that estimates the location and orientation of a robotic rover platform. The visual odometry approach is based on Fourier-Mellin transforms and phase-only matched niters which extract similarity transform parameters (i.e. rotation, scaling and translation) between consecutive images captured by a ground-pointing camera. By mapping these inter-image frame parameters to the real world, the rover's location is recovered. Given that the proposed strategy is a global method, it is well suited for environments exhibiting few features and illumination changes. Results show that this approach is more robust and accurate than an optical flow based visual odometry method.