Fusion of discrete and continuous epipolar geometry for visual odometry and localization

Abstract

Localization is a critical problem for building mobile robotic systems capable of autonomous navigation. This paper describes a novel visual odometry method to improve the accuracy of localization when a camera is viewing a piecewise planar scene. Discrete and continuous Homography Matrices are used to recover position, heading, and velocity from images of co-planar feature points. A Kalman filter is used to fuse pose and velocity estimates and increase the accuracy of the estimates. Simulation results are presented to demonstrate the performance of the proposed method.