Robust concolor object tracking against illumination change using PCA in color space

A novel robust Mean-shift tracking algorithm against illumination variation is proposed. The objects treated in this study are assumed to have single surface color. The real RGB color vector distributions of the object under various illumination conditions tend to concentrate on a specific line which is known as ¿color line¿ in the 3-D color coordinate. By applying the principal component analysis (PCA) to this kind of color distribution, less sensitive component against illumination condition is obtained as the third eigenspace. Therefore, the mapping each color data on the third pricipal axis gives color histogram with the least variation against illuminations. The proposed algorithm is applied to real image sequences containing objects under abruptly illumination change. The method can effectively track objects even for the cases in which the conventional Mean-shift and CAM-shift methods fail to track. Although the algorithm can be implemented in smaller amount of calculation, it has a challenge to handle in illumination changes.