KLI

Novel and Practical Object Re-identification and Search in Intelligent Surveillance Systems

Metadata Downloads
Abstract
Some of the most populated cities in all over the world are under a heavy amount of public surveillance systems for monitoring the population surrounding. In order to save labor costs, intelligent surveillance systems have rapidly developed in recent years by supplying and assisting security workers in detecting, analyzing, and predicting undesirable incidents.

Object re-identification (re-ID) and search are the foundation of a wide range of applications in intelligent surveillance systems. The targets of Object re-ID and search systems indicate person and vehicle. It can be used for a cross-camera person or vehicle tracking and search. The work on this manuscript focus on the camerabased object re-ID and search system in public datasets and real-world scenario.

To design a more realistic and practical object re-ID and search system for intelligent surveillance systems, we focus on three aspects. Firstly, this manuscript focus on investigating the unsupervised object re-ID. Secondly, we argue that training a system which able to identify a specific object from full scene images is closer to the real-world applications, therefore we investigate object search systems. Third, this manuscript focus on combining the supervised detection methods and unsupervised object re-ID methods.

We improve the performance of the re-ID systems by designing a more robust sampling strategy, refining pseudo labels, and designing loss functions. Moreover, we improve the performance of the search by designing learning strategies for unlabeled data and designing loss function. Extensive experimental results demonstrate the effectiveness of the proposed methods and their practicality in real-world unsupervised person re-ID applications. The experimental results of object re-ID have been evaluated on three public person re-ID datasets and one public vehicle re-ID dataset. The experiments are performed in two public person search datasets. Moreover, several outdoor real-world videos are used to validate the performance of the proposed methods in real-world applications.
Author(s)
당청
Issued Date
2022
Awarded Date
2022-08
Type
dissertation
Keyword
object Re-identificationobject searchperson re-identificationsupervised learningunsupervised learningsemi-supervised learningintelligent surveillance systemsconvolution neural network
URI
https://oak.ulsan.ac.kr/handle/2021.oak/10041
http://ulsan.dcollection.net/common/orgView/200000640946
Alternative Author(s)
Qing Tang
Affiliation
울산대학교
Department
일반대학원 전기전자컴퓨터공학과
Advisor
조강현
Degree
Doctor
Publisher
울산대학교 일반대학원 전기전자컴퓨터공학과
Language
eng
Rights
울산대학교 논문은 저작권에 의해 보호 받습니다.
Appears in Collections:
Computer Engineering & Information Technology > 2. Theses (Ph.D)
Authorize & License
  • Authorize공개
Files in This Item:

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.