一种基于混合高斯的快速背景更新方法

Method of Quick Background Updating Based on Gaussian Mixture Models

针对传统混合高斯模型对场景的突然变化不能实时更新和RunningAvg更新算法中容易产生拖影的问题,提出了一种快速背景更新方法。首先建立混合高斯和RunningAvg两幅背景,基于它们的二值化差分图像DB获取变化区域;将DB与前景二值化图像FB进行逻辑"与"以准确提取变化区域,消除拖影对目标提取的影响。然后根据变化区域的变化情况,用状态表中所记录的变化区域信息对背景模型的变化区域进行有选择的更新,减少了背景模型对变化区域的更新时间。实验结果表明,该方法不仅能对场景的突然变化具有很强的适应性,而且避免了拖影现象和物体短暂停留所造成的干扰。

The traditional mixture Gaussian models can not respond promptly to sudden changes in the background,and the smear phenomenon is inevitable in the RunningAvg update algorithm. Therefore ,an approach for quick background updating method is proposed. Firstly ,two backgrounds are constructed separately based on the Gaussian mixture models and the RunningAvg update algorithm. Then the binarized differential image DB of the two backgrounds is obtained to get the regions of variation in the scene. Logical ＂and＂ operation is utilized between the binary image DB and the binary fore- ground image FB in order to extract the regions of variation accurately,which can eliminate the negative impacts resulted from the smear phenomenon. Af- terwards,the information of the regions of variation,which is saved in the state table,is used to update the background models selectively depending on the changes in the regions of variation. As a result, the updating time of the regions of variation in the background models is reduced. The results have shown that the proposed method can not only strongly adapt to the sudden changes of the scene, but also can efficiently avoid the interference caused by the smear phenomenon and the short-term stay of the objects.