Precision multi-sensor optical navigation test-bed utilizing ground-truthed data set

Abstract

This work describes a rich set of navigation data that has been collected by AFRL and an object-oriented framework that makes both the data and relevant processing algorithms available to the research community. Both aerial and ground vehicle platforms were employed, exposing a single sensor suite to a variety of conditions relevant to civil and military applications, including urban and rural terrain and varied time-of-day. The sensor package includes four visible cameras, two LWIR cameras, inertial measurement units ranging from navigation grade, tactical grade and industrial grade. A GINS is utilized to establish platform attitude and position truth. The raw GPS signals are also collected. All this data is synchronously time-stamped with reference to a unified reference time. Of particular interest to the research community is the use of visual sensors to navigate in GNSS-limited environments, especially in cases where the navigation solution must exhibit dynamic continuity. Our data delivery framework goes beyond standardizing data formats by providing a means to standardize the application of relevant processing algorithms. This plays a critical role in integrating data from multiple sensors without violating dynamic constraints. In addition, the framework can be viewed as a testbed for user contributed algorithms, which will help system integrators to explore the trade space of both sensors and algorithms.