|Moment Matched Gaussian Kernel and Region Representative Likelihood for Performance Improvement of PMF-based TRN
Chang-Ky Sung* and Sang Jeong Lee
International Journal of Control, Automation, and Systems, vol. 18, no. 7, pp.1691-1704, 2020
Abstract : This paper proposes two techniques to improve the performance of Point Mass Filter (PMF) based Terrain Referenced Navigation (TRN). The first is the kernel generation technique for diffusion operation of the probability in the time propagation phase of the PMF, which matches the moments of the Gaussian distribution considering the length of the kernel. The second is a technique for calculating the likelihood for a representative region of a grid instead of calculating the likelihood at a grid in the measurement update stage of the PMF. Applying the proposed methods, it is possible to solve the overconfidence tendency problem of the conventional PMF and improve the stability of the filter, especially when the grid resolution is limited to a certain level or more because of the calculation amount. TRN simulation based on an interferometric radar altimeter (IRA) is performed to verify the effectiveness of the proposed technique and its results are described.
Moment matched Gaussian kernel, point mass filter (PMF), region representative likelihood, terrain referenced navigation (TRN)