Robust and illumination invariant change detection based on linear dependence for surveillance application

The subject of this paper is to provide an illumination invariant moving object detector for indoor surveillance applications. Furthermore we want to treat the illumination change as a mathematical/physical transformation procedure on images. Therefore the intention to this paper could also be defined as to provide an operator, which is invariant to transformations. The proposed detector relies on a model assigning a vector to every pixel location of the reference and the current image. The vector represents information on the neighborhood region of that pixel. Based on the above definition, the theorem of linear dependence of vectors is used to describe an operator for the detection of objects. For the purpose of an objective evaluation, the proposed technique is compared to the state-of-the-art Statistical Change Detection method. The proposed operator proved to be robust to noise as well as global illumination changes and local shadows and reflections.