ONTOLOGY BASED AUTOMATIC VIDEO ANNOTATION
Due to the rapid growth of the amounts multimedia information, the issue of its cataloging, analyzing, as well as search and retrieval, has become a challenging task. To solve this problem the process of video annotation is used.
This article provides a brief analysis of existing systems and the decision is made about the viability of using the ontologies in the form of hierarchically structured vocabulary of semantic video concepts and their basic properties as a tool to describe the contents of a video.
We propose a structure of the automatic video annotation system and then go into detail about each of its subsystems. An improved background subtraction method is used for moving objects segmentation, with the subsequent extraction of the objects’ features.
An ontology for the domain of human behavior analysis is developed. The ontology defines the basic set of events and scenarios, which is then used for the purpose of annotating the video sequences to enable semantic search and retrieval in video content. A Bayesian network to classify the events and scenarios in the videos is constructed using the domain ontology.
Finally, preliminary testing of the proposed system is carried out using the CAVIAR test video dataset and the results show great promise.