作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 图形图像处理 • 上一篇    下一篇

多相机网络中行人的全局最优匹配

吉培培,陈恳,郭春梅,李萌   

  1. (宁波大学 信息科学与工程学院,浙江 宁波 315211)
  • 收稿日期:2016-01-11 出版日期:2017-01-15 发布日期:2017-01-13
  • 作者简介:吉培培(1989—),女,硕士研究生,主研方向为计算机视觉、图像与视频处理;陈恳,副教授;郭春梅、李萌,硕士研究生。
  • 基金资助:
    国家科技重大专项(2011ZX03002-004-02);宁波市自然科学基金(2014A610065);宁波大学学科建设项目(XKXL1308)。

Globally Optimal Matching of Pedestrians in Multi-camera Network

JI Peipei,CHEN Ken,GUO Chunmei,LI Meng   

  1. (Faculty of Information Science and Engineering,Ningbo University,Ningbo,Zhejiang 315211,China)
  • Received:2016-01-11 Online:2017-01-15 Published:2017-01-13

摘要: 针对多相机视域下行人目标匹配正确率不高的问题,基于无监督显著性学习和局部特征匹配提出一种全局最优匹配模型。将不同视域间的目标匹配进行关联,每对相机的直接匹配受制于其间接匹配的监督,同时修正直接匹配中发生的误配。将经过亮度补偿后的图像帧分成若干局部块,通过无监督显著学习得到图像块的显著性得分,并结合目标图像块特征匹配的相似度得分,利用双向相似度计算目标间的相似度得分,并将其作为模型输入。基于标准数据库WARD和Shinpuhkan2014进行实验,结果表明,该模型能有效提高多相机监控网络下目标匹配的正确率。

关键词: 多相机网络, 特征融合, 全局最优匹配, 二值整数规划, 无监督学习, 亮度补偿

Abstract: In order to improve the pedestrian target matching accuracy under multi-camera horizons,this paper presents a globally optimal matching model based on unsupervised significant learning and local feature matching.It associates object matching between different horizons.Direct matching of each pair of camera is enslaved to the supervision of their indirect matching,and some errors in direct matching are corrected.The image frame is divided into a number of local blocks after brightness compensation,and significant score of image block is obtained through the unsupervised significant learning.Combined with the similarity scores of target image block feature matching,the similarity score between two objects is calculated by using bi-directional similarity calculation,and regarded as the input of the model.Experiment is based on two standard databases:WARD and Shinpuhkan2014,and the result shows that the proposed model can improve the object matching accuracy under multi-camera surveillance network.

Key words: multi-camera network, feature fusion, globally optimal matching, binary integer programming, unsupervised learning, brightness compensation

中图分类号: