改进的混合动静态背景的分割方法

An Improved Subtraction Algorithm of Backgrounds with Stationary and Non-Stationary Scenes

针对动静态背景场景下背景分割虚警率高的问题,提出了一种基于块直方图特征的Zivkovic混合高斯模型改进算法—BHZ-MoG。该算法设计了图像的块观测向量,并根据块观测向量的统计规律将图像块分类为动态、半动态、静态块,由此给出了结合块观测向量与块分类的块直方图特征提取算法,同时结合块直方图特征与Zivkovic混合高斯模型对不同类型的块分别进行背景分割与模型更新。实验结果表明,相较于Zivkovic混合高斯模型,BHZ-MoG算法能够在保证查全率不变的情况下,有效提高背景分割结果的查准率;Zivkovic混合高斯模型及BHZ-MoG的最大F1分数分别为0.758和0.790,说明了BHZ-MoG算法可以获得较佳的前、背景分割效果。另外,BHZMoG算法还可有效降低Zivkovic混合高斯模型在动态背景下的虚警率。

An improved algorithm--BHZ-MoG of Zivkovic-mixture of Gaussians （Z-MoG） based on block histogram feature is proposed to solve the problem that the background subtraction has a high false alarm rate in the mixture of stationary and non-stationary scenes. Observation vectors for image blocks are designed and blocks are classified into static, dynamic and half-dynamic blocks according to the statistical regularities of the observation vectors. A method that combines the observation vector and the classified information of a block is presented to extract block histogram feature. Block background models are constructed and updated from the combination of Z-MoG and histogram features. The BHZ-MoG can effectively reduce the high false alarm rate of Z-MoG under dynamic backgrounds. Experimental results show that the precision of the BHZ- MoG is higher than that of Z-MoG while the recall keeps the same. The maximal Fl-scores of the Z-MoG is 0. 758 and that of the BHZ-MoG is 0. 790, and it shows that the proposed algorithm can provide better subtraction results.