实时图像的运动目标检测及跟踪  

Moving Object Detection and Tracking for Real-Time Video

作　　者：安博文[1] 艾燕[1] 

机构地区：[1]上海海事大学信息工程学院,上海200135

出　　处：《计算机仿真》2012年第2期249-252,共4页Computer Simulation

基　　金：上海教委创新重点项目(11ZZ142);上海曙光计划(08SG49);上海自然科学基金资助(11ZR1415200)

摘　　要：在复杂背景的运动目标实时检测算法的研究中,由于目标受到外界环境影响,目标不能正确提取。针对克服背景干扰因素提取,干净的目标像素,大多数背景建模与背景更新算法计算复杂,难以满足视频监控的实时要求。为解决上述问题,提出一种根据像素特征的背景差法,将目标的边缘特征融入减背景算法,通过对离散的目标边缘梯度像素进行网格密度聚类法实现目标像素的提取,采用改进的均值漂移跟踪算法,在DM642平台上实现目标检测与跟踪。实验结果表明,改进的算法可以有效的克服光线变化、背景抖动、噪声等问题,实时检测、跟踪多个目标,并能解决目标遮挡问题。Background subtraction arithmetic is one of the practical and efficient moving objects detection algorithms based on still and complicated background, whose difficulty is how to get rid of noises and extract pure object pixels. Most of the background models need update using plenty of complicated calculation, which makes the real - time system hard to run. Aiming at the problems, the paper presented a new approach of background subtraction based on pixels features of objects. The method moved object edge information into background subtraction and then did object pixels clustering based on gridding density, and then tracked objects using the improved Mean Shift arithmetic. Experiments on DM642 show this method can detect and track multiple targets even when they cross each other under complicated and bad background.

关 键 词：边缘梯度 背景差 网格密度聚类法 均值漂移 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]