Title
Event Classification using Concepts
Author
de Boer, M.H.T.
Schutte, K.
Kraaij, W.
Publication year
2013
Abstract
The semantic gap is one of the challenges in the GOOSE project. In this paper a Semantic Event Classification (SEC) system is proposed as an initial step in tackling the semantic gap challenge in the GOOSE project. This system uses semantic text analysis, multiple feature detectors using the BoW model, SVM-based concept classifiers, event classifiers and fusion to classify if an event is present in a certain video. The TRECVID Multimedia Event Detection task 2013 is used to evaluate the SEC system. The results show that an initial step in bridging the semantic gap and tackling the challenges in the GOOSE project is made, but that there is room for improvement. We expect that future research in learning and defining high-level concepts and event classification will further bridge the semantic gap.
Subject
TS - Technical Sciences
Physics & Electronics ; Communication & Information
Information Society
Infostructures
Semantic Text Analysis
Event Classification
Feature Detection
Video
II - Intelligent Imaging ; MNS - Media & Network Services
To reference this document use:
http://resolver.tudelft.nl/uuid:85160446-10bc-4493-9cdc-dbf4b4a565cf
TNO identifier
485249
Source
ICT.OPEN 2013 The interface for Dutch ICT-Research, 27-28 November 2013, Eindhoven, The Netherlands
Document type
conference paper