基于自适应邻域的核密度动态目标分割方法

Method for dynamic object segmentation with kernel density estimation based on self-adaption neighborhood

经典的核密度估计背景模型使用固定的背景样本邻域来抑制背景运动形成的伪目标,无法适应不同背景的运动规律,导致不能抑制同一拍摄场景中所有背景运动形成的伪目标。因此在经典核密度估计的背景建模基础上,使用图像配准技术,能实现对不同运动背景区域的邻域尺寸自适应选择,并且在同一拍摄场景中可适应更多的背景运动类型,抑制更多类型的伪目标。实验结果证明,该方法对大部分由背景运动导致的伪目标有很好的抑制作用。

In classical kernel density estimation, using displacement probability with a small fixed window size to suppress faketarget, which caused by moving background. But fixed window size can not adapt to different movement of background. This pa-per designed and implemented a novel method based on background modeling with kernel density estimation. This method couldachieve window size＇ s adaptive selection. In the same filming scene, it could adapt to more moving background type and sup-press many kinds of fake target. Also the method used the technology of image registration. Experiment shows that this methodcan effectively suppress mostly false detection which caused by moving background.