National and Kapodistrian University of Athens
A multimedia content modeling and classification methodology using visual information for the protection of sensitive user groups
Abstract
dc:descriptionThe thesis concerns the problems of visual tracking and violence detection in video sequences. For the visual tracking problem, two feature fusion frameworks are presented. The first tracking framework called 'Model Fusion via Proposal' (MFP) framework, provides a way to efficiently fuse visual cues using independent trackers to construct an improved proposal distribution for the main tracker. The fusion method results in reduced computational requirements due to the better proposal and the gradual exploitation of the state space. The 'Hierarchical Model Fusion' (HMF) framework, extends the MFP framework by integrating the multiple models into a single tracker which exploits all the visual cues. This way the robustness of the approach is further increased. The frameworks are implemented and tested in various challenging sequences and proved robust with significant reduction of the computational cost. For violence detection, a system that classifies movie segments as violent or non-violent is proposed. The system fuses audio and visual information. The audio module uses state-of-the-art methods. The visual features concern the general motion in the scene, the detection of gunshots, and the motion of the detected people. The method is tested in a dataset comprised of segments from several movies.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Makris, Alexandros
- Μακρής, Αλέξανδρος
Subjects
dc:subject × 14- Φίλτρα σωματιδίων
- Οπτική παρακολούθηση
- Φιλτράρισμα
- Εντοπισμός βίας
- Οπτικοακουστικός συγκερασμός
- Particle filters
- Visual tracking
- Bayesian filtering
- Violence detection
- Audio-visual fusion
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23750