为了改善混合高斯模型在光照突变时容易产生大量误检的缺陷, 采用了一种高斯模型与均值法相结合并为前景像素建立计数器的方法。在建立背景模型时, 运用多帧图像求平均值的方法初始化混合高斯模型的背景;为每帧图像的前景像素数建立计数器, 并以此消除被误判为前景的区域;对检测出的前景区运用数学形态学处理, 得到图像真正的前景区域。结果表明, 该算法不仅克服了初始背景中的干扰, 而且消除了光照突变时的误检, 提高了运动目标的检测率。

In order to eliminate the defects of false detection of mixed Gaussian model under sudden illumination, a new algorithm combining Gaussian model with average background method was proposed to count the foreground pixels. Firstly, the background of Gaussian mixture model was initialized by using multi-frame averaging method when building the background model. Secondly, a counter for the number of foreground pixels of every frame was established and the false detection was eliminated based on the counter. Finally, the target was detected by using mathematical morphology and the foreground of the image was gotten. The results show that this improved algorithm not only overcomes the interference of the initial background but also eliminates the false detection when the illumination changes, and improves the detection rate of the moving targets.