Object tracking in thermal imaging using kemelized correlation filters

Object tracking algorithms in thermal imaging are very important in both civil and military applications, especially in the domain of surveillance and monitoring. Compared with visual object tracking, thermal infrared tracking has several advantages, but also possesses certain undesirable properties which makes the tracking task more challenging. With the aim of overcoming them, we apply a state-of-the-art Kernelized Correlation Filters (KCF) based tracking algorithm, which demonstrates high accuracy and robustness in visual tracking tasks, with very low computational complexity, and possibility of using almost arbitrary feature representations of the target, as well as nonlinear kernels. The algorithm has been tested on several chosen examples, using two different feature representations (raw pixels and HOG features). The simulation results are promising, with good performance even in the situations considered to be hard for thermal tracking.