A multistage target tracker in IR image sequences

Abstract

In this paper, we propose a robust approach for tracking target in forward looking infrared imagery taken from an airborne moving platform. The tracker is proposed in a two stage algorithm with a coarse to fine manner. First, we introduce a coarse but quick algorithm to roughly locate the target, which based on a model matching procedure by computing the SIFT vector of the target. The raw position and the size of the target are then used to initialize the multistage tracking algorithm. Second, the target’s position is refined through a dedicated localization algorithm, which based on a HOG feature tracking procedure to overcome the rapid size and angel changes of the dim target in the IR image sequence. Finally, the target’s model can automatically been updated and utilized in the first stage. As confirmed by experimental results on a variety of IR image sequences, the proposed approach efficiently and robustly tracks target under challenging environmental conditions. Moreover, the speed of the multistage algorithm is also convincing in the tested IR imagery.