Adaptive Background Modelling for Image Sequences with Cluttered Background
Maodi Hu, Yü Liu, Yiqiang Fan · 2014
Background subtraction is a key technique for video analysis applications. However, the existing algorithms do not work well in cluttered environments. In this work, we manage to model the oscillating background by using multi-channel background model, which is constructed by Gaussian filters with different variances. By employing a boosting-like updating rule for channel selection, a evidence-driving Adaptive Background Modelling (ABM) framework is proposed to eliminate false foreground responses. The effectiveness of ABM in tree and water regions is proven by experiments.