计算机与现代化

• 数据库与数据挖掘 • 上一篇    下一篇

面向物流仓库的多目标视频跟踪

  

  1. 河海大学计算机与信息学院,江苏南京210098
  • 收稿日期:2014-06-06 出版日期:2014-10-10 发布日期:2014-11-04
  • 作者简介:练海晨(1990),男,江苏南通人,河海大学计算机与信息学院硕士研究生,研究方向:计算机视觉,模式识别; 蒋亚平(1990),女,湖北孝感人,硕士研究生,研究方向:计算机视 觉,模式识别。

Multiobject Video Tracking in Logistics Warehouse

  1. College of Computer and Information, Hohai University, Nanjing 210098, China
  • Received:2014-06-06 Online:2014-10-10 Published:2014-11-04

摘要:

为了解决物流仓库复杂环境下多目标跟踪的问题,本文提出一种融合了背景建模的Camshift算法,并在算法跟踪过程中加入目标运动信息。首先根据跟踪目标获得目标颜色概率密度图像;然后根
据背景建模获得的背景图像对概率密度图像进行滤波处理;在Camshift每次迭代过程中,引入运动信息,通过加权融合获得最优位置;在多目标跟踪过程中,将当前帧中已跟踪完成的目标在概率密度图
像中滤除,减少对其它目标的干扰。通过实验表明,在物流仓库运送轨道上的物品跟踪的实际应用中,本文算法对复杂背景干扰和相似目标的相互干扰,有很好的处理能力。

关键词: 物流视频, 多目标跟踪, Camshift, 加权融合

Abstract:

 A Camshift algorithm, which is combined with the background modeling and the movement information, is presented in this article in order to resolve the multiobject
tracking problem in the complex environment of industrial field. Firstly, the matrix of probability density is calculated based on histogram of color of each object. Secondly,
these matrixes of probability density would be filter processed according to background image which is obtained from the background modeling. Thirdly, the movement information
is added to its iterative process, and the best position should be calculated with the weighting fusion. In the last, the Camshift removes the goals which have been tracked
successfully in the previous frame from those matrixes of probability density, in order to avoid a disturbance to other goals in the multiobject tracking. The experiment
indicates that this algorithm has a very good capacity of reducing background disturbance and similar goals disturbance.

Key words: logistics video, multiobject tracking, Camshift, weighting fusion

中图分类号: