Continuous Visual Navigation - An Evolution of Scene Matching

R. J. Handley, J. P. Abbott, C. R. Surawy

Abstract:	Accuracy and robustness are vital to present and future military airborne navigation. Despite considerable achievements in the field of inertial sensors and systems, raw Inertial Navigation System (INS) position accuracy is insufficient to support high speed Nap Of the Earth (NOE) flight profiles which are essential to ensure survivability and mission effectiveness of aircraft and Stand off Missiles (SOMs). In recent times this accuacy has been overcome by the use of the Global Positioning System (GPS), which provides accurate horizontal position and velocity data world-wide. The fusion of an INS with a GPS receiver, via a Kalman Filter, provides a very effective integrated system but suffers from poor vertical accuracy under normal conditions, and from degraded horizontal accuracy during periods of GPS unavailability. This eventuality is still a likely occurrence even with the enhanced robustness that is provided by a Controlled Radiation Pattern Antenna (CRPA). Terrain Contour Navigation (TCN) systems offer lower horizontal accuracy than GPS, but significantly higher vertical accuracy. However, over smooth terrain, such systems perform poorly. For many years, scene matching systems have offered horizontal accuracy equivalent to that of military (Precise Positioning Service, PPS) GPS, but have been unpopular due to their reliance on way-points, and the likelihood of becoming lost due to a bad fix. However, over the last few years, DERA and Hi-Q Systems have developed a technique which overcomes these historical weaknesses. This technique, called Continuous Visual Navigation (CVN), is a novel evolution of scene matching, and has been developed around the perceived way in which a human navigatcr might perform visual navigation: i.e. a combination of feature prediction and feature matching. The method removes the need for pre-planned way-points. In addition, in order to be robust to bad fixes, CVN utilises multiple parallel Kalman Filters which permit it to hedge its bets when there are several potential matches for a given feature. From the outset, design and development of CVN was concentrated on these novel aspects in order to avoid the pitfalls of traditional scene matching research (e.g. constantly ‘tweaking’ the image processing algorithms in order to improve peformance with a given data set). Consequently, although extremely effective, the image processing techniques employed are relatively simple. In addition, each CVN Kalman Filter is currently confined to model only four inertial error parameters. Even so, the system’s novel and clever design, which will be described in detail later in this paper, ensures that maximum use is made of all available featute data. This makes it very robust by scene matching standards. Currently, the CVN algorithms are being adapted to run in real-time, having previously only run off-line. The hardware platform for this adaptation will be a Pentium PC. The resulting system is scheduled for flight trials on a DERA aircraft in October 1998, and will navigate over a variety of terrain types in order to enable a broader study of the system’s capabilities.
Published in:	Proceedings of the 1998 National Technical Meeting of The Institute of Navigation January 21 - 23, 1998 Westin Long Beach Hotel Long Beach, CA
Pages:	217 - 224
Cite this article:	Handley, R. J., Abbott, J. P., Surawy, C. R., "Continuous Visual Navigation - An Evolution of Scene Matching," Proceedings of the 1998 National Technical Meeting of The Institute of Navigation, Long Beach, CA, January 1998, pp. 217-224.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In