自动化应用2024,Vol.65Issue(1) :5-8.DOI:10.19769/j.zdhy.2024.01.002

基于图像识别和数据分析的中厚板冷区钢板跟踪方法

Tracking Method for Steel Plates in Cold Zone of Medium-Thick Plate Factories Based on Image Recognition and Data Analysis

宋扬 陈星 张新昊 刘淼 戴林俐
自动化应用2024,Vol.65Issue(1) :5-8.DOI:10.19769/j.zdhy.2024.01.002

基于图像识别和数据分析的中厚板冷区钢板跟踪方法

Tracking Method for Steel Plates in Cold Zone of Medium-Thick Plate Factories Based on Image Recognition and Data Analysis

宋扬 1陈星 1张新昊 1刘淼 1戴林俐1
扫码查看

作者信息

  • 1. 北京首钢自动化信息技术有限公司,北京 100041
  • 折叠

摘要

现有的多摄像机多目标跟踪方法不能很好地解决网络延迟、目标遮挡、图像模糊等问题,为此,提出了一种基于图像识别和数据分析的多摄像机多目标跟踪方法.该方法采用双流网络的结构,在输入部分增加空间流和时间流,数据分析使用物理规则约束和修正目标行为,能在复杂环境下对跟踪区域内的钢板进行全程实时跟踪,且在钢板消失重现后,也能对钢板进行重定位.通过使用物理规则约束和修正目标行为,消除了网络延迟、目标遮挡、图像模糊对多摄像机多目标跟踪的影响.

Abstract

The existing MCMT(multi-camera and multi-object tracking)methods cannot effectively solve problems such as network delay,target occlusion,and image blur.Therefore,a MCMT method based on image recognition and data analysis is proposed.This method adopts a dual stream network structure,adding spatial and temporal flows in the input part,and using physical rules to constrain and modify target behavior for data analysis.It can track steel plates in the entire area in real-time in complex environments,and can also reposition steel plates after they disappear and reappear.By using physical rules to constrain and correct target behavior,the impact of network delay,target occlusion,and image blur on multi-camera and multi-target tracking has been eliminated.

关键词

多摄像机多目标跟踪/图像识别/深度学习/数据分析/钢板跟踪

Key words

MCMT/image recognition/deep learning/data analysis/steel plate tracking

引用本文复制引用

出版年

2024
自动化应用
重庆西南信息有限公司

自动化应用

影响因子:0.156
ISSN:1674-778X
参考文献量11
段落导航相关论文