A multi-criteria model for robust foreground extraction

Abstract

Numerous methods are currently available for motion detection using background modeling and subtraction. However, there are still many challenges to take into account such as moving shadows, illumination changes, moving background, relocation of background objects, and initialization with moving objects. This paper provides a new background subtraction algorithm that aggregates the classification results of several foreground extraction techniques based on UV color deviations, probabilistic gradient information and vector deviations, in order to produce a single decision that is more robust to those challenges.