Vision-Based Target Geo-Location Using Feature Tracking

A method for determining the location of a fixed ground target when imaged from the air using a camera equipped Micro Air Vehicle (MAV) is developed. The ground objects’ elevation is assumed known. Rather than a “one shot” affair, multiple bearing measurements of the ground object taken as the aircraft flies around the target are used. This makes it possible for the adverse effects of both the random measurement errors and the systematic measurement errors, a.k.a., attitude sensor biases, to be mitigated. The main result of this paper is that when bearing measurements of a target are taken over time, the target’s position and the MAV’s systematic attitude measurement errors can be jointly estimated using linear regression. As a result, the target is accurately geo-located and the attitude sensors are calibrated. If only one target bearing measurement can be taken, the attitude sensors will be calibrated prior to arrival at the target area using an Initial Point (IP). The technique has been successfully tested on actual flight data.