New infrared object tracking algorithm based on SMOG model

In color based object tracking, spatial-color mixture of Caussians (SMOG model) outperforms the popular color histogram in object discriminative power since it considers not only the colors in a region but also the spatial layout of these colors. In this study, SMOG model was applied to represent infrared objects, and the original similarity function in SMOG was revised in order to further improve its discriminative power. In particle filter framework, the two order AR model was utilized to describe the state transition equation, the revised similarity function was used as measuring to adapt the change of object appearance, and a valid model update method under this model was designed, then based on these, an efficient infrared object tracking algorithm was proposed. Experimental results show that the SMOG model can well represent the infrared object, and our proposed algorithm is effective and steady for tracking infrared objects.