Motion Imagery Navigation Using Terrain Estimates

Air vehicles rely heavily on inertial and satellite sensor feedback for navigation. Sensors such as gyroscopes and global positioning systems can provide accurate vehicle state information, but in many real world situations, these sensors can drift or bejammed, leading to navigation errors that can be especially problematic when operating near terrain features. In this proposal, we present a navigation system capable of estimating a 6-degree of freedom state vector purely from optical sensor input. Thesystem, Motion Imagery Navigation Using Terrain Estimates (MINUTE), is capable of estimating the state (position and orientation) of an unmanned air vehicle (UAV) given a (optional) initial state estimate, an existing terrain elevation map, and an imagesequence generated by an onboard sensor. This system first reconstructs a relative terrain map from pairs of images taken by a single camera on the UAV as it flies over terrain. The reconstructed terrain map is then matched against existing elevation datato produce an accurate, geo-referenced estimate of the state of the vehicle. In Phase I, we will design a working MINUTE prototype, study algorithm sensitivity using an existing synthetic data generation package, and address the real-time systemrequirements. Optical-based navigation state estimation is relevant to any autonomous system moving through a complicated environment. The technology developed could be used for Unmanned Ground Vehicles, Unmanned Aerial Vehicles (e.g., LOCAAS), orobserver systems where a human is in the loop controlling the vehicle.