Visual based Cloth Invariant Person Re-identification

报告人：张鹏 博士

单位：山东科技大学

题目：Visual based Cloth Invariant Person Re-identification

时间：2021年5月14日，下午2:30

地点：稼先楼南108

摘要：Person re-identification (re-ID) has been attracting extensive research interest because of its non-fungible position in applications such as surveillance security, criminal investigation and forensic reasoning. Existing works define it as a short-term event, which assumes pedestrians pass across surveillance network in a short-time period without changing clothes.  However, this assumption is not always true in practice, which pedestrians might re-appear after a long-time period. This task is termed as cloth invariant person re-identification.Besides challenges in classical person re-ID,cloth invariant person re-ID suffers severe within-person appearance inconsistency caused by clothing changes. These variations seriously degrade the performance of existing re-ID methods. To address the problem, we investigated methods from view of simulating cloth variation using vector capsules, exploring true motion pattern based on trajectory description and learningspatial-temporal information by a two-stream neural network, respectively. Experiments show the effectiveness of our efforts and demonstrate the feasibility of cloth-invariant person re-ID.

报告人简介：

张鹏，博士，分别于2013年、2016年获山东大学学士学位和硕士学位， 2020年获悉尼科技大学博士学位。2020年入职山东科技大学，主持并参与了包括国家自然科学基金在内的多项省部级项目，发表论文20余篇。主要研究方向包括计算机视觉、度量学习及深度学习等，相关研究成果发表在IEEE TIP、IEEE TMM、PR、IEEE TCSVT、WACV等国际知名期刊和会议，获授权发明专利10余项，受邀担任IEEETIP、IEEETMM、IEEETCSVT、IEEESPL、ICLR 2021、ICME 2020、ICML 2020等顶级期刊和会议审稿人。