Effects of foreground detection on probabilistic occupancy map tracking systems

Abstract

When there are serious occlusion and illumination change, Probabilistic Occupancy Map (POM) algorithm can track six more people simultaneously with multiple video streams. The output result of foreground detection will be the available input data in POM algorithm, so the performance of a foreground detection algorithm is very important for the entire POM algorithm. Presently, Gaussian mixture model (GMM) is a popular method for foreground detection. Visual background extractor (ViBe) improved by morphology algorithm can suppress the influence of the ghost that caused by shadows quickly and effectively. By using these foreground extraction algorithms to generate the foreground image, some error analysis experiments are carried out to explore the influence of foreground detection algorithm on the POM tracking algorithm in this paper. The experimental results show that POM can handle serious occlusion of multiple video sequences effectively and the morphology of ViBe improved algorithm outperforms GMM.