基于直方图灰度归类的快速背景建模方法

Fast Method for Background Modeling Based on Histogram Intensity Classification

针对传统核密度估计背景建模在检测精度和计算消耗上的不足,提出了一种基于直方图灰度归类的快速背景建模算法。首先根据序列图像像素灰度的相似性原理对背景进行直方图灰度归类,将邻近像素空间相关性转化为灰度直方图概率分布的关系,获得的直方图灰度与稳定背景相减剔除背景样本中的显著运动信息,减小了参与密度估计的样本数。然后根据背景分布特性选择最优窗宽,引入背景直方图概率进行加权高斯核密度估计,并通过自适应阈值实现运动目标的准确检测。同时,采用直方图背景更新,有效克服了光照变化对背景重建的影响。实验结果表明,该算法在检测精度和实时性处理方面都得到了很大提高。

In order to overcome the defects of accuracy and computational consumption for traditional kernel density estimate,a fast algorithm based on histogram intensity classification is presented for background modeling in this paper.Firstly,according to pixels ＇ similarity of sequence, the relations between adjacent pixels in spatial space were represented by probability distribution of histogram, and significant moving information were removed from gray information of histogram by background subtraction to reduce samples of density estimate. Then according to the characteristics of background distribution,the optimal bandwidth and weighted function were applied for gauss kernel density estimation. Moreover,the moving target could be detected accurately by adaptive threshold. Meanwhile,a histogram background updating method was applied,which overcame the light change on the background of the impact of the reconstruction. The experimental results showed that the algorithm greatly improve the accuracy of the foreground object detection and the speed of the detection processing.