Visual-inertial sensor fusion with a bio-inspired polarization compass for navigation of MAVs

Abstract

We present the integration of a polarization compass in a visual-inertial sensor fusion framework onboard a Micro Aerial Vehicle (MAV). The polarization compass estimates the position of the sun indirectly from the pattern of skylight polarization even in cases where the sun is not visible. It is based on a polarization sensor which consists of a standard RGB camera and a small polarizing unit that creates three polarization images on the camera sensor. Due to its low weight and compact size it is ideally suited for small aerial systems. The readings from the polarization compass are fused with angular rate and acceleration measurements from an Inertial Measurement Unit (IMU) and the 6 Degrees of Freedom (DOF) pose changes from the Visual Odometry (VO) in an indirect extended Kalman filter (EKF). Two different approaches to integrate the readings from the polarization compass in the filter are presented and compared. We show in experiments that adding a compass to visual-inertial sensor fusion does not only eliminate the drift of yaw angle estimates but also improves overall state estimation of the system.