摘要
针对智能视频监控场合对视频运动目标定位的需求,本文提出了一种基于非参数核密度估计的视频运动目标空域定位技术。该技术先对代表视频运动目标的前景样本点进行非参数核密度估计,选择具有最高密度指标的样本点为第一个目标中心,然后通过修正样本点的密度估计值,逐步实现对视频运动目标的空域定位。本文的方法是减法聚类视频运动目标定位技术的进一步推广。推广后的定位方法,可根据具体的目标定位场合,灵活选择核函数对样本点进行核密度估计。实验表明,本文方法具有良好定位效果,同时,在样本点的密度估计上更具灵活性。
To satisfy the need of video moving object locating in intelligent video surveillance scenes,video moving object locating technology based on nonparametric kernel density estimation is proposed.Nonparametric density estimation operation is firstly used on each foreground sample point that stands for video moving objects,and the sample point which has maximum density values is chosen as the first object center.And then other positions of video moving objects are gradually located by modifying the density value of sample point.This object locating method based on kernel density estimation is a further development of subtractive clustering object locating method.Using this method,kernel function could be flexibly chosen to estimate sample point's density values according to different locating application scenes.Experiment results show that the proposed method has much more flexibility in sample point's density estimation with satisfying locating results.
出处
《光电工程》
CAS
CSCD
北大核心
2010年第8期12-18,共7页
Opto-Electronic Engineering
基金
浙江省重大科技专项(优先主题工业项目)资助项目(2008C13076)
校科研基金项目(KYS055609080)
关键词
非参数密度估计
核密度估计
视频目标定位
视频目标检测
nonparametric density estimation
kernel density estimation
video object locating
video object detection