A Foreground Detection Method Based on Cross-Covariance Subspace Estimation

In this paper, a novel background modeling method was proposed based on cross-covariance subspace estimation to detect foreground in complex scenarios. The cross-covariance based 2DPCA (2 dimensional principal component analysis) method can preserve more image covariance information, which makes it suitable for background modeling. Therefore the cross-covariance based 2DPCA method was introduced into background modeling field and a correlative incremental algorithm was proposed for adaptively estimating background. Considering the sparsity and the continuity of the foreground, the method was used in foreground detecting accurately. Quantitative experimental and qualitative analysis results show that the proposed method can estimate the background information accurately and robustly in complex scenarios.