改进的核密度估计目标检测方法

Improved kernel density estimation method for target detection

为了增强核密度估计目标检测算法在实时监控系统的应用性能,在提高检测效果的同时减少运算量,提出一种改进的核密度估计目标检测算法。对原核密度算法进行深入分析,指出其原始样本是造成算法运算量大的主要原因;在此基础上,提出一种基于典型采样与多样性权值的改进核密度估计目标检测算法。提出更加灵活的样本更新方法,可随着背景变化快速更新样本信息,抗背景扰动效果明显。针对算法基于样本的特点提出一种基于样本的阀值分割方法,该方法能更好的与核密度估计算法融合,从而提高检测效果。通过实验验证了该算法的实时性和有效性。

To strengthen the kernel density estimation object detection algorithm in the application of real-time monitoring and control system performance,while improving the effect of detection of the computation,an improved kernel density estimation of the object detection algorithm was proposed.First of all,the original kernel density algorithm was analysed deeply,and a view was pointed out that algorithm large computational complexity was caused by the original sample of the proposed algorithm.Based on this,a kind of improvement kernel density estimation object detection algorithm based on the typical sampling and the diversity weights was advanced.A more flexible sample update method was proposed,which changed with the background and quick updated sample information.The effect of resistance background disturbancet was obvious.According to the characteristics of the algorithm based on the sample,a kind of threshold segmentation method based on the samples was proposed.This method could better integrate with kernel density estimation algorithm,thus improved the detection effect.The algorithm was verified to be real-time and effective by a real-time video.