基于核心密度估计的动态目标分割改进模型

数字视频中的动态目标分割是基于计算机视觉技术的分析、识别系统中关键的处理步骤,分割结果的正确率决定了后期分析或识别过程的质量.在数字视频中存在各种有害动态像素,它们会降低分割过程的正确率.动态目标的阴影是这些有害像素的一种,因为它们本身不属于动态目标形状信息的范畴,但分割过程却能很容易将它们作为有效的动态目标像素分割出来.最近有关动态目标分割的研究提出了基于核心密度估计模型的分割方法.基于RGB颜色空间的核心密度估计模型能够在彩色视频中抑制阴影,但是对于灰度视频这种模型是无法抑制阴影的.作者针对上述局限提出了一种基于像素边缘测量的核心密度估计模型,用于动态目标分割,能够在彩色和灰度视频中抑制阴影.实验结果证明,这种新模型在通常的应用条件下是有效的.

The Improved Model Based on Kernel Density Estimation for Dynamical Object Segmentation

Dynamical object segmentation in digital videos is an important step for many analysis systems or recognition systems based on computer vision technology.The accuracy of the segmentation result decides the quality of the following analysis or recognition procedures.There are kinds of harmful dynamic pixels which exist in digital videos and can deteriorate the accuracy of the segmentation process seriously.The cast shadows of dynamical objects are one kind of harmful pixels because they don't belong to valid objects'shape information,but the segmentation procedure may segment them for valid information easily.Recent researches of dynamical object segmentation bring forwardmethods based on kernel density estimation model.The kernel density estimation model for RGB color space can suppress the cast shadows in color videos,but it is not valid for intensity videos.In this paper,for these limitations,we provide a new kernel density estimation model based on pixel edge measurement to segment dynamical objects,suppressing cast shadows in both color videos and intensity videos.And,some experiment results are presented and they prove this model is valid on normal conditions.