Aristotle University Of Thessaloniki (AUTH)
Μέθοδοι μηχανικής μάθησης και μηχανικής όρασης για την ευφυή ανάλυση εικονοσειρών
Abstract
dc:descriptionIn this dissertation, the results of research conducted in the areas of intelligent video analysis using machine learning and computer vision methods are presented. Although emphasis was placed on movie/TV production data, in order to showcase the potential of modern artificial intelligence in media production and post-processing industry, the proposed algorithms can be more generally applied to any video type. The presented research considers the problems of stereoscopic quality defects detection, human activity recognition in stereoscopic 3D videos, automated stereoscopic movie summarization conforming to narrative properties and automated human activity video summarization. Our main contribution to the problem of stereoscopic quality defects detection consists in describing four algorithms for automatically detecting and characterizing corresponding types of defects, in the post-processing stage of movie/TV content production. Regarding the problem of human activity recognition in stereoscopic videos, ways to exploit scene geometry information encoded by the stereoscopic disparity channel are proposed, with the goal to augment human activity recognition performance in natural settings. Our research was extended to the problem of automated, multimodal stereoscopic 3D movie summarization conforming to narrative properties, in the form of a video skim. To this end, a complete algorithmic summarization pipeline was developed that considers visual, audio, geometric and narrative characteristics in movie shots and video frames. Finally, the issue of automatically summarizing long activity videos was studied, which share a number of common, recurrent properties (static camera, static background, high degree of inter-frame visual similarity) and may be derived from a variety of sources (surveillance cameras, shooting sessions in movie/TV production, etc.). To solve the problem, a novel algorithmic framework was developed for summarizing activity videos, in the form of a representative key-frame set that optimally sums up the various depicted activities.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Mademlis, Ioannis
- Μαδεμλής, Ιωάννης
Subjects
dc:subject × 16- Μηχανική όραση
- Μηχανική μάθηση
- Αναγνώριση προτύπων
- Ανάλυση εικονοσειρών
- Computer vision
- Machine learning
- Pattern recognition
- Video analysis
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY_NC_ND
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44837