TIB’s Visual Analytics Research Group to develop methods for person detection and visualisation
DFG funds new joint research project for automatic indexing in scientific video archives
HANNOVER, 19 March 2018 – This January saw the start of the project “Development of a software system for automatic scene and person indexing in scientific video archives “, which is being funded by the German Research Foundation (DFG) for a period of three years. The partners involved in the project are the German Broadcasting Archive (DRA, Director Angelika Hörth), the Chair of Practical Computer Science at Philipps-University of Marburg (Professor Bernd Freisleben) and the Visual Analytics Research Group (Professor Ralph Ewerth) of the Technische Informationsbibliothek (TIB) – German National Library of Science and Technology – Leibniz Information Centre for Science and Technology.
Within the project, the University of Marburg will develop methods for the detection of concepts and scenes, whilst TIB will concentrate on approaches for person detection and the visualisation of results. The German Broadcasting Archive (DRA) will contribute to the project from a documentary perspective by conducting requirement analyses and a qualitative evaluation. In addition, it will make available historic TV material on a large scale – at its Potsdam-Babelsberg site, the DRA manages the entire wealth of television programmes of the former German Democratic Republic (GDR), and is the only point of contact for scientific research in this field. The project partners will be able to build on the excellent results of a forerunner project.
One of the first goals is to further optimise detection rates for concepts and persons and, with the use of the envisaged deep learning methods, to expand these to other research-intense parts of DRA television broadcasts. In addition, the project will involve the implementation of the automatic indexing of material for 100 further visual concepts and 100 further important persons of the GDR, enabling video content-based search enquiries to be conducted on this basis. Another objective is to ultimately enable archive staff to initiate training processes for the underlying machine learning methods for adding further concepts or persons. The resulting software system should enable the DRA to respond quickly and flexibly to different research questions placed by archive users. However, the research project does not only seek to develop a sustainable solution for the entire material of the DRA, it is also supposed to be easily transferable to other archives. For this purpose, the software tools developed within the project will be made available in open source, enabling other research institutions with media archives to benefit from the person lexicon, concept lexicon and visualisation techniques, and to adapt them to their own needs with little effort.
Professor Ewerth is looking forward to advancing the content detection methods in videos and the visualisation of results: “As a result, we hope to be able to provide other institutes and organisations with reusable open source software tools for automatically indexing their video archives, contributing to the indexing of video data relevant to research.”
Acting in the capacity of the German national library of technology, as well as architecture, chemistry, computer science, mathematics and physics, the TIB –Leibniz Information Centre for Science and Technology and University Library provides academia, research, industry and business with literature and information in printed and electronic form.
TIB is continuously expanding its role as a German information centre for the digitisation of science and technology. The library provides scientific content, digital services and methodological skills to specialist and research communities at www.tib.eu, supporting the different stages of scientific work. The library’s search and order portal offers users access to more than 80 million technical and scientific datasets indexed by TIB. The library’s outstanding quality-assured collections also include knowledge objects such as audiovisual media, 3D models and research data. Thanks to specially developed search technologies, TIB’s AV-Portal (https://av.tib.eu/) can be used to search for specific content from scientific videos from the fields of science and technology. As a research library, TIB conducts applied research and development in order to generate new services and optimise existing ones. The key areas of research are in the fields of data science, non-textual material, Open Science and visual analytics.
TIB is a public-law foundation of the Federal State of Lower Saxony. The library is a member of the Leibniz Association.