Recursive State Estimation of Nonlinear Systems
2001 (English)In: Proceedings of the Third Conference on Computer Science and Systems Engineering, 2001, 175-181 p.Conference paper (Refereed)
Particle filters (sequential Monte Carlo methods) handle recursive state estimation of arbitrary systems. However, a direct application shows that often a vast number of particle is needed for the filter to work well. This paper developes a combined particle and Kalman filter for recursive state estimation of nonlinear systems, where the state vector can be partitioned into one linear and Gaussian part and one nonlinear and/or non-Gaussian. The linear and Gaussian part is estimated using the Kalman filter and there maining part is estimated using the particle filter. Based on Bayes rule the solutions from the two filters are blended together. This two-filter approach is compared to a standard particle filter by applying both filters to a simple navigation problem. The result shows that using the combined filter the number of particles needed to achieve the same performance as the particle filter is much less.
Place, publisher, year, edition, pages
2001. 175-181 p.
Particle filter, Kalman filter, State estimation, Nonlinear, Navigation
Engineering and Technology Control Engineering
IdentifiersURN: urn:nbn:se:liu:diva-90780OAI: oai:DiVA.org:liu-90780DiVA: diva2:616485
Third Conference on Computer Science and Systems Engineering, Norrköping, Sweden, March, 2001