Implicit sampling for particle filters

Area of Science:

• Computational statistics
• Nonlinear filtering
• Particle methods

Background:

• Particle-based nonlinear filtering is crucial for estimating probability density functions in complex systems.
• Existing methods often require a large number of particles, increasing computational cost.
• Chainless Monte Carlo methods offer alternative approaches to particle filtering.

Purpose of the Study:

• To develop a novel particle-based nonlinear filtering scheme.
• To improve the efficiency of particle filtering by reducing particle requirements.
• To enhance the accuracy of probability density function estimation.

Main Methods:

• A particle-based nonlinear filtering scheme is proposed.
• Each probability density function is represented by a set of Gaussian variable functions.
• Resampling is performed using normalization factors and Jacobians.

Main Results:

• The proposed scheme focuses particle paths sharply.
• Fewer particles are required for accurate estimation.
• The method is demonstrated on an ill-conditioned test problem.

Conclusions:

• The novel filtering scheme offers improved efficiency and accuracy.
• The method effectively reduces particle count in nonlinear filtering.
• This approach shows promise for complex estimation problems.