Clutter adaptive tracking of multiaspect targets in IRAR imagery

Mixed-state particle filters for multiaspect target tracking in image sequences

“…If N p << L2, the computational savings may be considerable. Overall, our simulations suggest that the particle filter trackers compare favourably to the grid-based HMM tracker in [6] by yielding similar RMSE performance roughly 95 % of the time, hut at a much lower computational cost …”

“…The initial target velocity in both dimensions is a sample from a Gaussian random variable with mean 10mls and standard at instants n = 0 and n = 9, with peak target-to-clutter ratio (PTCR) equal to 3.6 dB. We tracked the simulated target over 13 consecutive frames using respectively the SIR filter in Table l , the APF filter in Table 2, and the grid-based multiaspect HMM tracker introduced in [6]. Both the SIR filter and the APF use N p = 5: 000 particles.…”

“…We see from the plots in Figure 2(b) that the auxiliary particle filter has a slower target acquisition time, but eventually converges to a steady-state RMSE that is indistinguishable from the error of the HMM filter. The HMM tracker is however computationally expensive requiring at each frame, see [6], the evaluation the likelihood function in all points of the image grid for all aspect states. For an L x L grid, that means a computational cost of order 0 ( d 2 )…”

“…In section 3 , we present the SIR and APF mixedstate trackers. In section 4, we examine the performance of the proposed algorithms and compare them to the pointmass HMM tracker previously introduced in [6]. Finally, we present our conclusions in section 6.…”