Sequential Monte Carlo filtering for multi-aspect detection/tracking
Marcelo G. S. Bruno, R.V. de Araujo, Anton G. Pavlov · 2005
We propose in this paper a mixed-state sequential Monte Carlo (SMC) filter for joint multiframe detection and tracking of multiaspect targets in cluttered image sequences. The proposed detector/tracker is a sampling/importance resampling (SIR) particle filter that uses resampling according to the weights to combat particle degeneracy and also includes an additional Metropolis-Hastings (MH) move step to avoid particle impoverishment. The dynamic models for target motion and target aspect and the statistical model for the spatially correlated background clutter are assumed as prior knowledge in the design of the filter. The performance of the algorithm is investigated using simulated image sequences generated from real infrared airborne radar (IRAR) data.