Precise Estimation of Geo-location Uncertainty

ABSTRACT: As unmanned aerial vehicles have become equipped with high-quality cameras, the warfighter has gained access to large quantities of video and motion imagery with both wider areas of coverage and finer resolutions. For these data to be truly useful, the warfighter needs to be able to precisely geo-locate any pixel in the imagery. As no geo-location algorithm will be perfect, the warfighter also needs an accurate measure of the confidence in the geo-location solution. Under this effort we will build on our Phase 1 approach for modeling errors in image-based geo-location algorithms that enables more accurate estimates of precision. We will develop an integrated model for accurately capturing uncertainties in multiple, related error sources of image processing-based geo-location algorithms, resulting in unified framework for estimating the geo-location uncertainty. We will implement real-time algorithms for estimating the model parameters, and evaluate the model against the current state-of-the-art in terms of the accuracy of the uncertainty estimates. The resulting modeling and inference approach will enable real-time, accurate estimates of image-based geo-location and its uncertainty. BENEFIT: As the military employs geo-registration across a number of applications, such as image-based navigation and targeting, the technology developed on this program will benefit a wide range of programs. The accurate characterization of geo-location uncertainty will provide a measure of confidence in the results from geo-registration algorithms, enabling the fusion of geo-location information with other sensors. Specifically, for image-aided navigation, this geo-registration information can be fused to enable accurate navigation in GPS-denied environments. Because the approach builds on existing computer vision algorithms, it has direct commercial application to virtually any autonomous robotic deployment scenario, like emergency search and rescue and remote exploration.