Back to search

Aristotle University Of Thessaloniki (AUTH)

Μηχανική όραση για ανίχνευση γεγονότων σε βίντεο στοχαστικών κινήσεων

Abstract

dc:description

This dissertation presents the development of innovative techniques aiming at handling video scenes, as captured by static surveillance cameras for security purposes. Specifically, the problem of anomaly detection, defined as divergences from dominant paths, is studied in-depth for a variety of real word situations, in an effort to prevent potential crisis events. In this work, different methodologies concerning three widespread categories in video surveillance (crowds, traffic and dynamic textures) are proposed in order to successfully tackle all the problems arising in the area. Firstly, a newly introduced concept based on swarm theory is applied for the extraction of robust motion characteristics. Swarm intelligence is exploited and, in combination with appearance features, contribute to the form of a powerful descriptor, effectively describing crowded scenes’ dynamics. At the same time, a hybrid local-global approach related to the extraction of multiple hierarchical Dirichlet processes is adopted in order to deal with traffic video sequences. The proposed method successfully captures scene’s correlations and rules, often found in these kind of scenarios, leading to an efficient traffic analysis. Finally, a novel scheme, focusing on detecting and localizing a wide range of dynamic textures, as captured by surveillance cameras, is introduced. More precisely, a new handcrafted descriptor is adopted, while a framework based on multi-scale superpixel clustering is used in order to detect textures of high interest for outdoor surveillance (e.g. fire, water etc.). In each case, the proposed scheme is validated in a wide range of scenarios showing a remarkable performance and proving our method’s generality and applicability in real world situations. This claim is also supported by the low computational cost displayed by all algorithms, making them very appropriate for a variety of surveillance applications.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2018

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kaltsa, Vagia
  • Καλτσά, Βάγια

Subjects

dc:subject × 10

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/43753

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kaltsa, Vagia; Καλτσά, Βάγια. Μηχανική όραση για ανίχνευση γεγονότων σε βίντεο στοχαστικών κινήσεων. Aristotle University Of Thessaloniki (AUTH), 2018. http://hdl.handle.net/10442/hedi/43753