Foreground detection based on co-occurrence background model with hypothesis on degradation modification in dynamic scenes

摘要

This work presents a Hypothesis on Degradation Modification (HoD) based on Co-occurrence Pixel-Block Pairs (CPB, which is proposed in our previous work) to further resist background changes for foreground detection, such as illumination changes and background motion. HoD provides CPB with a model update strategy that can be used for a long time. While further improving the robustness of CPB, it also stabilizes the efficiency of CPB over time. A key contribution of this work is it offers a robust background subtraction for foreground detection in dynamic scenes. The observation is robust to illumination changes and background motion and demonstrates the ability of HoD. Experimental results obtained from the datasets under different challenges of PETS 2001, AIST-Indoor, SBMnet and CDW-2012 databases prove that our algorithm has a good effectiveness for foreground detection. (C) 2019 Elsevier B.V. All rights reserved.