前沿讲座202204：Research on Key Issues from Pedestrian to Group Retrieval under the Cross-View Video Surveillance

报告题目：Research on Key Issues from Pedestrian to Group Retrieval under the Cross-View Video Surveillance

报 告 人：Ling Mei, Sun Yat-sen University

报告时间： 2022年4月14日16：00

报告地点：南一楼中311

主持人：俞耀文

摘要：Group retrieval mainly studies the cross-view identification of the group that is composed of several pedestrians, which is of great importance in the field of public security applications, and deep learning based artificial intelligence (AI) techniques have promoted its developments. However, there are two main challenges: 1) Most previous studies focus on single pedestrian re-identification (re-id) and ignore the correlations among group members, and they lack a large and comprehensive group retrieval benchmark to associate these two tasks. 2) The illumination variation in surveillance is another challenge to obtain stable optical flow estimation for group motion, and current cross-view group retrieval researches mainly consider the asynchronous issues, which ignores the temporal synchronization and spatial overlap in the practical surveillance. To solve the above challenges, in this talk, we will go over the identification and motion analysis from pedestrian to group retrieval under the cross-view surveillance. Specifically, this talk consists of two parts: On the one hand, for the identification issue, we will cover how to build a large and comprehensive benchmark to associate the re-id and the group retrieval, and extract robust and consistent representations of group retrieval features by designing state-of-the-art deep learning network, and optimize the traditional re-id task by the contextual group retrieval information. On the other hand, for the crowd motion analysis, we will cover how to get an efficient illumination-invariance optical flow estimation across different views, then use it to measure the crowd motion collectiveness via the global motion correlation. Finally, we will cover an unsupervised and domain-adaptive motion feature based video synchronization algorithm to alleviate the cumbersome group annotation problem in surveillance.

简介：Ling Mei has participated in a national scholarship supported visiting Ph.D. scholar project with the Department of Computer Science at the University of British Columbia (UBC), Vancouver, BC, Canada from 2020 to 2021. He received the Ph.D. degree in Information and Communication Engineering from Sun Yat-sen University, Guangzhou, China, in 2021, and the M.S. degree in Pattern Recognition and Intelligent System from Sun Yat-sen University, Guangzhou, China, in 2016. He has been an Intern at the National Key Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences (CASIA), Beijing, in 2016. He has published papers in international SCI journals and conferences on computer vision and artificial intelligence, e.g., the IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), Neurocomputing, and IEEE ICPR. He has served as the reviewer in the TCSVT journal. He received the International Program Award for Young Talent Scientific Research People of Guangdong Province in 2019, and has won twice the 2nd Prize of the first and second National Graduate Contest on Smart-City Technology and Creative Design in 2014 and 2015, respectively. His research interests include Group retrieval, Person re-id, Optical flow, Video synchronization, and Deep learning networks.