Natural Language Based Query Formulation for Video Retrieval Using Spatio-Temporal Query
No Thumbnail Available
Date
2017-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
High speed processors and large storage devices have made videos popular and easily available to everyone. With this dynamic growth, different research areas are created with content based video indexing, grouping, searching and retrieval. Video is processed based on the content of visual features and annotated metadata which is required for the identification of features, events and objects in a video.
One of the main problems of retrieving information from database is that casual users may not be aware of the structure of the database. The idea of using natural language, instead of SQL has prompted the development of query processing to fill this gap. Using natural language casual users can retrieve information on video database without the help of experts. The formulated queries can represent the stated object and event of user‟s query by using different spatial, temporal and predicate operators. So it would be easy to map and query relevant video scenes while preserving user stated objects and operators.
In this study we proposed natural language based query formulation for video retrieval at video scene level. Natural language queries are preprocessed to have ready for query formulation process. The query formulation process creates a high level SQL form which holds all necessary information to be ready for query execution phase. The designed data model supports a spatio-temporal and predicate based query for complex user queries. In addition, if the stated object query is not found in database user can retrieve semantically similar objects using the similarity metrics algorithm. A prototype application has been developed using appropriate tools and techniques for the soccer video domain. The prototype application has been tested after seen the full video content and checked with different query types. The system has been found to be 78% accurate to return video scenes that match with the user queries.
Keywords: Natural Language Processing, Natural Language Interface for Databases (NLIDB), Query Formulation, SQL, Similarity Metrics, Video Retrieval
Description
Keywords
Natural Language Processing; Natural Language Interface for Databases (NLIDB); Query Formulation; SQL; Similarity Metrics; Video Retrieval