Ground Vehicle Navigation Based on the Skylight Polarization

Abstract

Autonomous or unmanned ground vehicles can take advantage of camera-based navigation systems. These navigation systems mainly rely on standard radiometric cameras. The use of polarization information, such as captured by a polarization filter array camera, is a potential extension to capture multimodal information efficiently. In this communication, we propose a navigation method that relies exclusively on Stokes images reconstructed from polarization camera data. For this purpose, an image processing pipeline is employed to estimate the heading of a vehicle. To assess the method, an acquisition card has been built and coupled with two moving platforms: a rotary stage and a moving ground vehicle. The results show that, in a dynamic car experiment, the root mean square error of the orientation is 4.29° as compared to a Global Positioning System/Inertial Navigation System.