摘要
针对不同相机视角间的域偏移问题,例如遮挡、光照、背景等域信息变化而引起的视觉差异,提出一种基于联合字典对学习的跨视角行人重识别算法。具体地,由于同一相机视角下的图像具有域相似性,因此通过低秩分解细化用于表示域信息的字典,即用一个字典表示相同相机视角下的域信息,而用另一个字典代表行人外观特征信息。该方法的分离思想主要是使来自同一相机视角下的所有行人图像在字典上享有相同的稀疏表示,以此分离每个视角下行人图像共享的域信息和行人特征信息。实验表明,所提方法的Rank-1值相较于次优算法在PRID2011、CUHK01和i_LIDS数据集上分别提高1.2%、1.8%和4.49%,识别性能与鲁棒性更优,以期为跨视角行人重识别提供参考与借鉴。
Aiming at the problem of domain offset between different camera views,such as visual differences caused by changes in domain information such as occlusion,illumination,background,etc.,a cross-view pedestrian recognition algorithm based on joint dictionary pair learning is proposed.Specifically,because images from the same camera perspective have domain similarity,the dictionary used to represent domain information is refined through low-rank decomposition,that is,one dictionary represents domain information from the same camera perspective,and another dictionary represents pedestrian appearance feature information.The main idea of this method is to make all pedestrian images from the same camera perspective have the same sparse representation in the dictionary,so as to separate the domain information and pedestrian feature information shared by pedestrian images from each perspective.Experiments show that the Rank-1 value of the proposed method is better than that of the suboptimal algorithm in PRID2011,CUHK01 and i_The LIDS data set is improved by 1.2%,1.8%and 4.49%respectively,and the recognition performance and robustness are better,so as to provide reference and reference for cross-view pedestrian recognition.
作者
颜悦
程清翠
李向奎
朱豪
YAN Yue;CHENG Qing-cui;LI Xiang-Kui;ZHU Hao(School of Physics and Information Engineering,Zhaotong University,Zhaotong 657000,China;Sichuan Fanxicheng Science and Technology Ltd,Chengdu 610213,China)
出处
《软件导刊》
2023年第5期198-205,共8页
Software Guide
基金
云南省教育厅科学研究项目(2023J1209)
云南省基础研究计划项目(202001AP040046)。
关键词
行人重识别
联合字典对学习
域信息分离
低秩分解
person re-identification
joint dictionary pair learning
domain information separation
low-rank decomposition
