摘要：AbstractA common technique for analyzing certain biophysical phenomena involves \tagging" a feature of interest with a uorescent particle and observing it over time. This technique, known as particle tracking microscopy, typically involves two steps: (1) localization of the particle in space, and (2) estimation of parameters related to a mathematical model of the particle's motion. There exist several methods that accomplish each task independently; however, there is currently no general framework that can perform both tasks simultaneously with arbitrary motion and observation models. In this work, we describe how this can be accomplished through the application of a framework recently developed by Schön, Wills, and Ninness. This framework, which uses the Expectation Maximization algorithm in conjunction with Sequential Monte Carlo methods, can simultaneously localize a particle's location while calculating maximum likelihood estimates of model parameters. Since the precision and computational complexity of this method significantly depend on the number of particles used, we also describe a potentially faster, albeit suboptimal, method based on the assumed normality of the posterior densities. We demonstrate these methods by applying them to the tracking of a nanometer-scale uorescent particle undergoing anisotropic diffusion in a plane.
