A robust moving object segmentation algorithm using integrated mask-based background maintenance
Yuyang Chen, Yanyun Zhao, Anni Cai · 2012
In this paper a novel moving object segmentation algorithm is proposed. This algorithm utilizes an integrated mask-based background (BG) maintenance scheme. A rough foreground (FG) mask and a block-based FG mask are integrated by an OR operation to give the final update mask. Multiple updating rates are used for different regions under different circumstances. Additionally, by analyzing the distribution of pixel values of the temporal difference, we provide an effective detection scheme for sudden illumination change, after which the BG model is rebuilt. Experimental results based on the challenging I2R dataset show that our proposed method provides better performance in comparison with other classic and state-of-art algorithms such as GMM and Codebook, both qualitatively and quantitatively. According to the overall metrics of F1 score, our method achieves 23.74% higher quality of FG segmentation than GMM.