Target tracking based on a hybrid tracker with two hierarchical appearance models
Fanglin Wang, Erqi Liu, Jie Yang, Ruiming Liu · Measurement Science and Technology · 2007
In this paper, a hybrid target tracking algorithm is proposed based on two hierarchical appearance models. First, the colour histogram is utilized to roughly localize the object by a mean shift procedure. Then, a more precise eigenspace appearance model was invoked to infer the final state of the object within a particle filter framework. To deal with the sudden illumination change, the histogram equalization algorithm was also used to prevent the eigenspace model from being incorrectly distracted. Moreover, a mean shift algorithm in affine space was designed to address the affine deformation of the tracked object. Numerous experiments show that the proposed algorithm performs well in the presence of significant appearance change, large illumination variations and partial occlusions.