Tracking ground based targets in aerial video with dual-tree wavelet polar matching and particle filtering

The detection and tracking of targets in aerial imagery of cluttered urban environments is addressed. Polar matching, using dual-tree complex wavelet transforms, is used as a shift and rotation invariant detector. A particle filter is employed to add robustness, especially in the event of target occlusion. We show that, together, these methods can robustly track a ground based target as it becomes both partially and fully occluded.