一种改进的基于背景抽取的运动目标检测算法  

Improved Method of Moving Object Detection Based on Background Subtraction

作　　者：杜向龙[1] 伍健荣[1] 邢涛[1,2] 

机构地区：[1]中国科学院上海微系统与信息技术研究所无线传感器网络与通信重点实验室,上海200050 [2]无锡物联网产业研究院,江苏无锡214135

出　　处：《计算机测量与控制》2011年第11期2665-2667,共3页Computer Measurement &Control

基　　金：国家科技重大专项(2010ZX03006-004)

摘　　要：视频监控针对的场景是安静的场景,但是具有随机的扰动,如树木摇动、杂物抖动等;针对这种场景,提出了对基于混合高斯模型的运动目标检测算法的改进方法,在混合高斯模型检测到运动目标的初步结果基础上,采用锐化处理、平滑处理、二值化处理等手段,保留图像固有特征,滤除随机抖动;对处理后的图像运用背景帧差法,弥补混合高斯模型的不足,最终检测到准确的运动目标;实验结果表明,该方法能从具有随机扰动的视频流中准确的检测到运动目标。The scene monitored by intelligent video is stable most of the time, but has random perturbation, such as the waving of trees and the shaking of litter. For this type of scene, a improved method for the moving object detection algorithm which is based on the Mixture of Gaussians has been proposed. For making up the deficiency of Mixture of Gaussians, this method can retain the inherent characteristics of the image and filter out the random perturbation by sharpening, smoothing and dichotomy processing based on the preliminary result of moving object detection generated by the Mixture of Gaussians and uses the background difference algorithm to detect the moving object finally. Experimental results indicated that the method can detect the moving object from the video stream precisely.

关 键 词：运动目标检测 混合高斯模型 锐化 平滑 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]