Global Tracking based Multi-Object Tracking in Complex Environments

Qin, Yulong

Global Tracking based Multi-Object Tracking in Complex Environments

Qin, Yulong

Permalink

Publication Type:: Thesis
Issue Date:: 2025

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (10.93 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Qin, Yulong
dc.date.accessioned	2025-09-22T23:54:42Z
dc.date.available	2025-09-22T23:54:42Z
dc.date.issued	2025
dc.identifier.uri	http://hdl.handle.net/10453/190016
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Multi-object tracking (MOT) is a core task in computer vision, widely applied in autonomous driving, surveillance, and human-computer interaction. However, it remains challenging in real-world scenarios due to occlusion, motion blur, and missed detections. To address these limitations, we propose a novel transformer-based MOT framework that enhances temporal continuity, attention modeling, and robustness. Specifically, we incorporate a Kalman filter prediction module to supplement unreliable or missing detections, ensuring uninterrupted trajectory generation. We further integrate the Convolutional Block Attention Module (CBAM) into the transformer to emphasize key spatial and channel-wise features. To better handle dynamic target motion, we introduce MQTFormer, a multi-query tracking strategy that leverages multiple trajectory queries across frames. In addition, a multi-weight fusion mechanism combines feature similarity, detection confidence, and adaptive weighting for more robust association. Extensive experiments on standard benchmarks, including MOT17 and TAO, demonstrate that our method outperforms state-of-the-art approaches in terms of accuracy, robustness, and stability, especially under challenging conditions.	en_US.UTF-8
dc.format	Thesis (MSc (CompSc) )
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/190016/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2025 Yulong Qin
dc.rights	au.edu.uts.lib/nph
dc.title	Global Tracking based Multi-Object Tracking in Complex Environments	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Multi-object tracking (MOT) is a core task in computer vision, widely applied in autonomous driving, surveillance, and human-computer interaction. However, it remains challenging in real-world scenarios due to occlusion, motion blur, and missed detections. To address these limitations, we propose a novel transformer-based MOT framework that enhances temporal continuity, attention modeling, and robustness. Specifically, we incorporate a Kalman filter prediction module to supplement unreliable or missing detections, ensuring uninterrupted trajectory generation. We further integrate the Convolutional Block Attention Module (CBAM) into the transformer to emphasize key spatial and channel-wise features. To better handle dynamic target motion, we introduce MQTFormer, a multi-query tracking strategy that leverages multiple trajectory queries across frames. In addition, a multi-weight fusion mechanism combines feature similarity, detection confidence, and adaptive weighting for more robust association. Extensive experiments on standard benchmarks, including MOT17 and TAO, demonstrate that our method outperforms state-of-the-art approaches in terms of accuracy, robustness, and stability, especially under challenging conditions.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/190016