Effects of Non-Gaussian and Non-White Noise Processes on Image-Based Targeting for Mission-Critical Applications

M.J. Veth

Abstract:	The revolution in sensor and computational technology has enabled the development of many novel techniques for extracting navigation and targeting information using natural signals. One noteworthy research area has focused on the use of imaging sensors as navigation sources. These image-aided navigation techniques have been experimentally demonstrated to provide errors on the order of, or sometimes better than, tactical-grade inertial sensors. This accuracy potential has resulted in a great deal of interest for military applications, especially on Intelligence, Surveillance, and Reconnaissance (ISR) platforms. Unfortunately, accuracy alone does not fully address the overarching requirement for navigation and targeting solution integrity in mission-critical applications. In a typical image-aided navigation system, a feature extraction algorithm is used to convert each image to a collection of features which are a projection of a location in the world called a landmark. The most desirable features are distinct and invariant to changes in the observation of the associated landmark. These features are then matched between two or more images to produce a putative matching set. Using projective geometric constraints, the match sets can be used to determine both the camera pose and the three-dimensional location of each landmark. While there are some variations in the algorithms presented in the literature, the most general technique is called bundle adjustment and is essentially a variable-dimensional nonlinear regression. The optimal solution to the bundle adjustment is highly dependent on the assumptions regarding the statistics of the measurements. In the large majority of cases, these errors are simply assumed to be stationary, white, and Gaussian, which is known to be inadequate for image-aided navigation measurements. This is complicated by the requirement to specify a putative match set which is known to be susceptible to correspondence errors. This results in a measurement set that is clearly non-stationary, non-white, and non-Gaussian in nature and can result in state estimates which are overly optimistic. In this paper, the error statistics for an image-aided navigation system are developed from first principles and are shown to be non-stationary, non-white, and non-Gaussian in nature. The effects of these errors are demonstrated on the navigation and targeting solution for various simulated unmanned aerial vehicle (UAV) trajectories. The resulting error statistics of the state estimate are analyzed and compared to those predicted by a naive error model. Conclusions are drawn and architectures for improving the integrity of image-aided navigation and targeting systems are presented.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	1988 - 1995
Cite this article:	Veth, M.J., "Effects of Non-Gaussian and Non-White Noise Processes on Image-Based Targeting for Mission-Critical Applications," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 1988-1995.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In