基于背景差分的核密度估计前景检测方法

摘要

针对非参数核密度估计算法前景检测不够精确、运算量大的问题,提出了一种基于背景差分图像的核密度估计前景检测方法.该方法结合了单高斯模型和核密度估计模型进行初始背景建模,利用背景差分图像,过滤掉非动态背景区域,对动态背景区域采用核密度估计进行像素分类.同时,对非动态背景区域,采用渐进式更新；对动态背景区域,采用非参数核密度估计进行更新.实验结果表明,该算法能够精确地分割出前景目标,减少了误检噪声,降低了运算量.%In view of existing situation that foreground object detection is imprecise and calculation is large in non-parametric kernel density estimation, the foreground object detection algorithm based on background subtraction image kernel density estimation is proposed. The single-Gaussian model and the non-parametric kernel density estimation are applied to building the initialization scene background. By means of background subtraction image, the non-dynamic background region is filtered. Kernel density estimation is used to estimate the motion object. For the dynamic background region, the algorithm uses non-parametric kernel density estimation algorithm to update it, otherwise, the percent of background and current frame is used to progressively update the non-dynamic background region. The experimental results show that this algorithm can separate foreground object accurately, greatly eliminate the false detection and reduce the calculated amount.