Road-constrained target tracking using particle filter

Abstract: In this work a particle filter (PF) that uses a one-dimensional dynamic model to estimate the position of vehicles traveling on a road is derived. The dynamic model used in the PF is a second order linear-Gaussian model. To be able to track targets traveling both on and off road two different multiple model filters are proposed. One of the filters is a modified version of the Efficient Interacting Multiple Model (E-IMM) and the other is a version of the Multiple Likelihood Models (MLM). Both of the filters uses two modes, one for the on road motion and one for the off road motion. The E-IMM filter and the MLM filter are compared to the standard PF to be able to see the performance gain in using multiple models. This result indicates that the multiple model filters have better performance, at least when the true mode switching probabilities are used.