Mean Shift Based Target Tracking in FLIR Imagery via Adaptive Prediction of Initial Searching Points

Tracking targets using mean shift algorithm is performed by iteratively translating a kernel in the image space such that the past and current target observations are similar. The standard mean shift, however, assumes that the initialization point falls within the basin of attraction of the desired mode. When tracking objects in FLIR imagery this assumption may not hold, particularly when the target´s displacement between successive frames is large. A novel mean shift based tracking algorithm based on adaptive prediction of initial searching points is proposed to make the tracker procedure converge to the global mode of the density function. The infrared target is represented in the cascade grey space. We use Kalman filter to track the variance of target coordinate transformation parameter. Especially, an on-line noise model estimation mechanism is developed to accommodate different target movement adaptively. Experiment results show our novel scheme is efficient and robust for the infrared targets with severe clutter background and large displacement.