Performance characterization of infrared imagers for horizon-matching geolocalization methods

Image-based geolocalization methods show promise as a means of augmenting GPS-dependent navigational systems that may be susceptible to intermittent signal loss. Among these, horizon matching is one effective approach for regions wherein terrain is topographically feature-rich. In this method, horizon lines are extracted from images of surrounding scenery, and these are compared to computationally generated horizons derived from a digital elevation database. By optimizing the vehicle position in the simulated space to reduce error mismatch the real and computed horizons lines (i.e., cost function minimization), vehicle position can be derived. We have implemented this approach in a mobile multi-camera system in order to compare performance of several spectral bands including the visible, near-infrared, short-wave, and long-wave infrared regions. Here, we report our progress in establishing metrics by which these different imaging bands may be evaluated specifically for horizon matching. Geopositioning accuracy is categorically important, but it is not the only relevant consideration. Understanding how camera attributes such as resolution and contrast-to-noise performance contribute to accuracy is critical for advancing these systems. Furthermore, the different bands are confounded by different lighting and atmospheric conditions which preclude horizon extraction. The implication is that that the ideal system will therefore utilize the bands synergistically, and this effort aims to investigate how to best do so.