A novel approach for event foreground detection based on PSO parameter optimization and region partition based background modeling

Abstract

This paper proposes a novel image sequences based approach for event foreground detection in wild scenes with strong environment noises, which bases on PSO parameter optimization and region partition. First of all, base on variance vectors of row and column of difference image, the background model is partitioned into estimated event region covering real event foreground and estimated background region. Then, specific optimal Kalman filters for background modeling are designed for the two regions respectively. At last, event foreground is detected by utilizing adaptive threshold based segmentation method based on the optimal background model. Experimental results demonstrate that the proposed approach could achieve a better performance, while it detects a more accurate event foreground with extremely low false results, by strengthening the event foreground and inhibiting the environment noises.