《计算机应用研究》|Application Research of Computers

基于辨识特征后融合的行人再识别

Discriminative feature based late fusion for person re-identification

免费全文下载 (已被下载 次)  
获取PDF全文
作者 刘琦,侯丽,彭章友
机构 1.上海大学 通信与信息工程学院,上海 200444;2.黄山学院 信息工程学院,安徽 黄山 245041
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2019)08-066-2552-04
DOI 10.19734/j.issn.1001-3695.2018.02.0186
摘要 跨摄像机行人因光照、视角、姿态的差异,会使其外观变化显著,给行人再识别的研究带来严峻挑战。基于多特征融合和距离度量学习技术,提出辨识特征后融合的算法,并将其应用于行人再识别中。对跨摄像机行人样本图像分别提取局部最大出现频次(LOMO)特征和基于显著颜色名称的颜色描述子(SCNCD)特征,表示跨摄像机行人的外观。基于所提取的LOMO和SCNCD特征,分别去训练跨视图二次判别分析(XQDA)距离度量学习模型,分别获取跨摄像机每对行人每个特征优化的距离;应用最小最大标准化距离融合的算法,获取跨摄像机行人最终的距离,用于跨摄像机行人的匹配。在具有挑战性的VIPeR和PRID 450S两个公开数据集上进行实验,实验结果表明所提出的行人再识别算法有效地提高了行人再识别的准确率。
关键词 行人再识别; 多特征融合; 距离度量学习; 距离融合; 最小最大标准化
基金项目 国家自然科学基金资助项目(61704161)
安徽省教育厅自然科学研究项目(KJHS2016B03)
黄山学院横向科研项目(hxkt20170059)
黄山学院校地合作项目(2017XDHZ021)
本文URL http://www.arocmag.com/article/01-2019-08-066.html
英文标题 Discriminative feature based late fusion for person re-identification
作者英文名 Liu Qi, Hou Li, Peng Zhangyou
机构英文名 1.School of Communication & Information Engineering,Shanghai University,Shanghai 200444,China;2.School of Information Engineering,Huangshan University,Huangshan Anhui 245041,China
英文摘要 Pedestrian may vary greatly in appearance due to differences in illumination, viewpoint, and pose across cameras, which can bring serious challenges in person re-identification. This paper proposed discriminative feature based late fusion using multiple-feature fusion and distance metric learning for person re-identification. Firstly, it expressed human appearance across cameras through local maximal occurrence(LOMO) and salient color names-based color descriptor(SCNCD) extracted from human sample images across cameras. Secondly, it obtained each optimized feature distance of each pair of pedestrians through individual cross-view quadratic discriminant analysis(XQDA) distance metric learning model trained based on LOMO feature and SCNCD feature, respectively. Finally, it obtained the final distance of each pair of pedestrians across cameras through a distance fusion using min-max normalization, which it applied for human match across cameras. Experimental results show that the proposed algorithm effectively improves the accuracy of person re-identification on two challenging datasets(VIPeR, PRID 450S).
英文关键词 person re-identification; multiple-feature fusion; distance metric learning; distance fusion; min-max normalization
参考文献 查看稿件参考文献
 
收稿日期 2018/2/12
修回日期 2018/3/21
页码 2552-2555
中图分类号 TP391.41
文献标志码 A