Videotorium – video sharing for Hungarian research and education

posted by Andras Kovacs on Thu, 11/25/2010 - 4:00pm

The whole story of recording and web publishing of recordings at NIIF/HUNGARNET (Hungarian research and education network, NREN) started back in 2003 when our yearly conference was first recorded and webcasted. As know-how emerged and other institutions saw its usefulness we were invited to record more and more material on scientific conferences and events. This led to the foundation of our “Video on Demand” archive that was built on a simple web interface and Windows Media (with so called WM "script" tags for slide control). The repository had been slightly improved with years (backend database, archive was converted to VP6/flash, etc.) to improve browser compatibility and easy of use, but could not fulfill the needs for user upload feature, modern navigation/retrieval and metadata description.

We started designing and implementing our new portal called "Videotorium" in 2009 with financial support from locally co-ordinated EU and national funds. The aim of the development was the following:

Web based media repository for research and higher-education,
Upload feature up to HD quality,
Metadata capabilities to meet scientific retrieval requirements,
Support of Shibboleth AAI,
Basic community features (commenting, rating, embedding, sharing, etc.),
Accommodate old material from “Video on Demand” repository,
Metadata aggregation features,
Live broadcast through the portal.

By target user groups the usage of Videotorium is not strictly limited to Hungarian higher-education and research community, any education related non-profit materials are welcome in the repository.

Architecture

Videotorium is built mostly on open source software, such as Apache/PHP, Lighttpd, MySQL database, Memcached, ffmpeg, mplayer, etc. The architecture is constructed from fairly independent components, which concept enables us to relocate any of these components to other servers when necessary.

As the portal supports many input formats for upload, a background conversion must take place before publishing the media in H.264 format ready for flash based web playback. All the media is serviced by Wowza Media Server, which allows flexible provisioning at good performance. RTMP streaming is preferred in the first place, but fallback to HTTP tunneled RTMP is also supported for strictly firewalled/proxied networks.

Metadata capabilities

We put significant effort to create a metadata model that allows an accurate retrievability of content. A group of experienced librarians were involved to work out a metadata structure that is a good compromise between:

Professional description of media content meeting scientific requirements,
Allowing easy metadata creation by non-specialist (i.e. non-librarian) uploaders.

Although, this task had not been very easy at all, but carefully designed user interfaces seemingly ease the problem. The metadata model itself is built on Europeana Semantic Elements (ESE), but includes a considerable extension by introducing three levels of metadata description:

Basic metadata (partly mandatory): recording title(s), subtitle(s), description, time of recording, list of contributors (names, jobs and organization) and their roles, contributor key pictures, genre, technical metadata, etc.
Science field classification (optional): based on Ortelius thesaurus once developed by the European Union for research application classification. Videotorium is using the dual language (Hungarian and English) version of Ortelius provided by its maintainer Budapest University of Technology and Economics National Technical Information Centre and Library (BME OMIKK).
Content description (optional): using free tags or the dual language (Hu/En) version of the famous Library of Congress Subject Heading (LCSH) thesaurus maintained by LoC and University of Debrecen University & National Library for translated version.

The metadata model is using an approach, where a dual language description is optionally possible for the uploader: original language (language of the recording) and English. This not only supported by metadata fields, but as well as by multi language thesauruses. This would allow finding non-English recordings using English keywords.

This concept of metadata model is currently being tested and might be revised based on user feedback and usability experiences.

Authentication

The authentication/authorization behind Videotorium supports both Shibboleth AAI (branded as eduID in Hungary) and "legacy" user/password. All static file requests (download of media files, slides, attached documents, etc.) requests are serviced through Lighttpd server after checking authorization information through an internal authentication service. The streaming server – using a Wowza plugin – is also integrated with this internal component. This solution allows a good security control over media playback and download.

Publishing process

The process of publishing a recording starts with upload of media file(s) and other documents:

Media file(s): original media file and further language versions (e.g. with translated audio track).
Upload of presentation to be synchronized to media (supported formats: PPT, ODP, SXI and PDF),
Upload attached documents, images, etc.

After successfully uploading media file(s) a background conversation takes place. Based on the resolution/quality of the original media file(s) different H.264/AAC surrogates are created (low/high quality). In addition, an audio only version is prepared and index thumbnails are also extracted from video. The index thumbnails are also offered as contributor key images, so uploaders can easily identify the particular contributor record to use with other recordings.

The uploaded presentation files are automatically converted to still images and thumbnails in the background, the content is full text indexed. Videotorium provides an intuitive flash application to manually synchronize slides to media by easily placing them on the timeline. This application also allows uploading still images one by one.

Videotorium editor

As a supporting feature, an automatic slide OCR subsystem has been developed to allow optical recognition of uploaded still image slide content. Thus, the slide content becomes searchable and allows viewer to jump to specific slides in media (see screenshot). The Videotorium OCR system is based on cuneiform open source OCR framework. This also helped us to make the previously captured 45,000 bitmap slides searchable.

The portal

The portal follows a structure similar to most video sharing portals. The navigation is possible using events/channels, science fields, featured recordings, etc. The media is played back using a specialized flash player showing resizable video/slide areas, and providing features such as fullscreen, media embed/share, slidelist, etc.

Videotorium player

Metadata aggregation features

Videotorium intends to become a national research and higher-education multimedia aggregator by harvesting organizational repositories using OAI-PMH protocol. Videotorium DC is specified that would be a common language for provisioning metadata to Videotorium. The portal will be also capable of pushing Videotorium metadata to other – higher level or international – aggregators.