视频流中的单向快速运动多目标可靠跟踪技术  

Accurate Tracking Technology for Fast Moving Targets in Video Stream

作　　者：王荣[1] 李强[1] 王骞[1] 冯楠[1] 孙娟[1] 

机构地区：[1]国家知识产权局专利局专利审查协作北京中心,北京100190

出　　处：《电视技术》2013年第S2期447-450,共4页Video Engineering

摘　　要：提出了一种基于扩展Kalman预测方法的多目标跟踪技术,首先寻找多个目标的主运动方向,并向其垂直方向进行灰度累积投影,将相邻两帧的灰度累积投影结果进行配准,就可得到多目标的整体位移估计量,为配准提供数据支撑。同时,对视频流中同一帧的目标进行提取并获取中心点坐标,针对这些坐标建立扩展Kalman预测模型,实现可靠的帧间目标位移量预测,再通过后续匹配定位处理进行目标的精确定位及跟踪。该方法适用于有较高实时性要求的大量单向运动目标跟踪应用中。在基于图像处理的棒材在线计数研究中,使用该方法,解决了在传送带上的多支棒材目标的定位和跟踪难题,在相邻帧棒材存在相对运动的极端情况下,跟踪成功率大于99%。A video based fast moving targets tracking technology is introduced here.For the great inter-frame displacement,inter-section,it is always a problem to perform in the field appliance.In order to solve this problem,an EKF based multi-targets tracking method is proposed to find the main movement direction,then the gray-level for interframe matching is accumulate to calculate the whole displacement.Then,an EKF chain is established,to give an accurate track for each target who come into the window.This method is suitable for fast moving target mainly in the same direction.Using this method,a tracking problem is solved in rebar plant for on-line counting and separating,error rate lower than 1%.

关 键 词：视频流 快速运动 目标跟踪 模板匹配 

分 类 号：TN919.8[电子电信—通信与信息系统]