Infrared Target Tracking with AM-FM Consistency Checks

Multichannel dual domain infrared target tracking for highly evolutionary target signatures

“…However, our work in [11,12] suggests strongly that the failure modes in the two domains are almost never the same. Thus, even when both domains fail simultaneously, it is extremely rare that the two simultaneous failures are similar.…”

“…If the template is updated every frame, we have found that the track filter invariably over adapts, ultimately losing the target and locking onto structured features of the clutter. While the fixed interval update strategy represents an improvement in this regard, it is plagued by the unsolved problem of how to choose the update interval L. In [11,12], we showed by a proof of concept experiment (using a normalized correlation tracker) that it is possible to detect when the template has become stale by observing the target simultaneously in two domains: viz., the pixel domain and the modulation domain.…”

Dual domain auxiliary particle filter with integrated target signature update

“…Template tracking in the modulation domain has also been demonstrated recently [11,15]. In practice, the target detection processes is often coupled with a tracking algorithm consistent with either a Kalman or Particle filtering methodology [2,3,17].…”

“…The problem of refreshing the target template is well known and is referred to in the literature as the template update problem [5,9,10,13]. Several strategies for preventing stale templates include updating the template every frame, updating the template at a fixed or variable time interval, and, more recently, updating the template based on track consistency checks which detect divergence of the tracked target centroid in the pixel and modulation domains [11,12].…”

“…The new dual domain track processor maintains three independent particle populations, two of which are used to detect when the target template has become stale and one which tracks the target jointly in the fused modulation and pixel domain feature space. In addition, we improve upon the recent dual domain track consistency checks used in [11,12], which rely on the distance between the pixel domain and modulation domain target centroids alone, by quantifying the difference between the target signature evolution in the two domains more comprehensively. Although we continue to threshold the distance between the modulation domain and pixel domain centroids, the auxiliary particle filter given here explicitly integrates the observed target appearance into the resampling step that is performed just prior to estimating the target size and location.…”