Target tracking in infrared imagery using a novel particle filter

To address two challenging problems in infrared target tracking, target appearance changes and unpredictable abrupt motions, a novel particle filtering based tracking algorithm is introduced. In this method, a novel saliency model is proposed to distinguish the salient target from background, and the eigenspace model is invoked to adapt target appearance changes. To account for the abrupt motions efficiently, a two-step sampling method is proposed to combine the two observation models. The proposed tracking method is demonstrated through two real infrared image sequences, which include the changes of luminance and size, and the drastic abrupt motions of the target.