基于Parzen核估计的动态背景建模算法

An Method for Dynamic Scene Modeling Based on Parzen Kernel Estimator

利用非参数估计理论核密度估计方法,研究了多模态场景参考帧的动态维护与更新问题。针对户外伴有树木晃动、水波荡漾等问题的复杂环境,在利用Parzen核估计算法进行动态背景建模的基础上,对该算法进行改进。首先,在运动前景检测阶段,给出了一种改进的阈值设定算法,提高了像素点分类的准确性。其次,在去除噪声阶段,利用像素邻域相关性,降低了目标检测的虚警率。实验结果表明:该算法可更有效地对运动目标进行检测,并具有较强的实时性。

A kernel density estimation （KDE） based on a muhimodal model is presented for dynamic scene reference frame maintenance and update problems. Aim at complex outdoor environment with small movements, such as tree branches moving with the wind or rippling water, an improved method based on Parzen kernel estimator is proposed for dynamic scene modeling. Firstly, in moving foreground detection, an improved threshold setting method is presented that can distribute pixels more accurately. Secondly, in noise removing, Pixel＇s neighbor information is considered to decreases the false positive in object detection. The experimental results indicate that the proposed approach in this paper can process in real-time and detect moving targets more effectively.