Rotation and Scaling Invariant Target Tracking Using Particle Filters in Infrared Image Sequences

Abstract

A new rotation and scaling invariant target tracking algorithm is proposed using particle filters. Specifically, the target aspect is modelled by a continuous-valued affine model which is augmented to the target's kinematic parameters and whose dynamics are assumed to follow a first-order Markov model. Two specific particle filtering algorithms are implemented, i.e., Sequential Importance Re-sampling (SIR) and Auxiliary Particle Filter (APF). The Gaussian-Markov Random Field (GMRF) is used to characterize the spatial clutter of the background, and a target signature model is used to simulate the presence of a target. Simulation results show good tracking performance on targets with time varying rotation angles and scale factors even under low signal-to-noise ratios.