Error event simulation for HMM tracking algorithms using importance sampling

Importance sampling is a technique for speeding up Monte Carlo (MC) simulations. The fundamental idea is to use a different simulation distribution to increase the relative frequency of "important" events and then weight the observed data in order to obtain an unbiased estimate of the parameter of interest. This estimate often requires orders-of-magnitude fewer simulation trials than ordinary MC simulations to obtain the same specified precision. We present an importance sampling technique applicable to error event simulation of hidden Markov model (HMM) tracking algorithms. The computational savings possible with the use of this technique are demonstrated both analytically and using simulation results for a specific HMM tracking algorithm.