多分辨率差分图像非参数核密度估计方法仿真

针对多分辨率差分图像核密度估计阶段中,由于信息冗余与重复计算导致的估计结果准确率下降的问题,提出一种非参数核密度估计方法。利用硬件设备采集多分辨率视频序列,提取关键帧图像作为样本集。分割多分辨率的差分图像,形成由背景图像与前景运动目标两部分组成的初始模型。以该模型为基础构建Copula核函数,利用核函数的运算性能分别确定估计窗宽、方差和核密度公式,从而输出差分图像非参数核密度的估计结果。通过仿真得出结论:研究方法平均准确率为98.56%,与传统核密度估计方法相比提升了6.04%,证明此方法具有较高的应用价值。

Nonparametric Kernel Density Estimation for Multiresolution Differential Images

The accuracy of estimation results is decreased due to information redundancy and repeated calculation in the stage of kernel density estimation for multi-resolution differential image.Therefore,a method of non-parametric kernel density estimation was put forward.Firstly,the hardware equipment was used to collect the multi-resolution video sequence and extract the key frame of image as the sample set.Then,the multi-resolution differential image was segmented to form an initial model which was composed of background image and foreground moving target.Based on this model,Copula kernel function was constructed and the calculation performance of kernel function was used to determine the estimation window width,variance and kernel density respectively,and thus to output the estimation result of non-parametric kernel density of difference image.Simulations show that the average accuracy of the proposed method is 98.56%,which is improved by 6.04%compared with that of traditional kernel density estimation method.