Improving Accuracy of MAV Pose Estimation using Visual Odometry

Abstract

We present a system for estimating MAV location and attitude with increased accuracy by coupling GPS/INS telemetry information with visual odometry (VO). An on-board camera provides image data from which VO information can be extracted, providing another source of information about aircraft pose. We present a technique for estimating and propagating the uncertainty associated with VO-based pose estimates, allowing this information to be fused with GPS-based estimates in an extended Kalman filtering framework. We present results demonstrating a substantial increase in accuracy of pose estimates.