Facilitating vehicle re-identification across distinct UAV data captures

Maniotis, Konstantinos - Marios; Μανιώτης, Κωνσταντίνος- Μάριος

dc.contributor.author	Maniotis, Konstantinos - Marios	en
dc.contributor.author	Μανιώτης, Κωνσταντίνος- Μάριος	el
dc.date.accessioned	2025-06-12T11:48:44Z
dc.date.available	2025-06-12T11:48:44Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/62054
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.29750
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Vehicle Re-identification	en
dc.subject	Aerial Surveillance	en
dc.subject	Transformer Models	en
dc.subject	Deep Learning	en
dc.subject	UAV Imagery	en
dc.subject	Επαναπροσδιορισμός οχήματος	el
dc.subject	Εναέρια επιτήρηση	el
dc.subject	Μοντέλα Transformer	el
dc.subject	Βαθιά Μάθηση	el
dc.subject	Εκόνες UAV	el
dc.title	Facilitating vehicle re-identification across distinct UAV data captures	en
dc.title	Δυνατότητα επαναπροσδιορισμού οχήματος σε διαφορετικές λήψεις δεδομένων UAV	el
heal.type	masterThesis
heal.classification	Computer Vision	en
heal.language	en
heal.access	campus
heal.recordProvider	ntua	el
heal.publicationDate	2024-10-12
heal.abstract	Vehicle re-identification (Re-ID) in UAV-captured aerial imagery poses challenges such as variations in perspective, scale, and occlusions. This research reproduces results from three established repositories—Relation Preserving Triplet Mining (RPTM), LuPerson for person Re-ID, and TransReID—before evaluating performance on a custom UAV-captured dataset, Suncity, representing real-world urban traffic. The proposed methodology includes the use of RPTM for optimizing triplet mining by respecting natural data groupings and TransReID, a transformer-based model designed to capture both global context and fine-grained features. Enhancements such as the Jigsaw Patch Module (JPM) to handle occlusions and the Side Information Embeddings (SIE) module to account for camera viewpoint and lighting variations further improve the model’s robustness. While the models performed well on benchmark repositories, including improvements in mean Average Precision (mAP) and Rank-1 accuracy, challenges were identified when applying the models to the complex Suncity dataset. Despite these challenges, transformer-based models demonstrated superior adaptability over traditional CNNs. This work contributes to advancements in vehicle Re-ID, with potential applications in traffic monitoring, surveillance, and urban planning.	en
heal.advisorName	Karantzalos, Konstantinos	en
heal.committeeMemberName	Karantzalos, Konstantinos	en
heal.committeeMemberName	Stamou, Georgios	en
heal.committeeMemberName	Voulodimos, Athanasios	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών	el
heal.academicPublisherID	ntua
heal.numberOfPages	75 σ.	el
heal.fullTextAvailability	false
heal.fullTextAvailability	false