In order to keep a low operational cost per channel, the production tools developed within the ICoSOLE project are targeted at commodity hardware. This poses several technical challenges that are related to the efficient implementation of processing many video streams in a real-time scenario as described by this project. We will investigate GPU-based methods to efficiently process video frames within a broadcasting scenario. There are several state of the art approaches for GPU video transfers that we would like to extend for the needs of a broadcasting-grade video processing pipeline.
The solutions found to address These challenges are not only relevant to the production tools of the ICoSOLE project, but also to other stages of ICoSOLE that operate within the domain of Video During production, UGC and professional content will continuously become available after they have passed the quality analysis and filtering and phase. We will investigate how the quality of the final programme delivery on the intended output channel(s) can be optimized by presenting a location-aware and timeline-based overview of all available and checked sources that can contribute to these output channels. This includes:
- support for appropriate blending of professional and UGC content
- taking into account visual similarity for UGC at the same stage or location
- bility to distinguish interesting from unwanted or non-interesting “incidents”
- alignment of audio-video sources on timeline
Another important innovation aspect of ICoSOLE is the metadata handling throughout this
new type of production process. Metadata will be captured and added at an early stage,
wherever possible via linked open data. There should also be an easy mean in case of UGC
contributions to annotate or appreciate the captured content. Together with the technically
acquired metadata via content analysis, all the gathered metadata will play an important role
in the semi-automatic (multi-platform) visualization of content sources and in the composition
of the final programme. It will also enable to create automatic summarizations of an event.
These summarizations can be both generic (assembled by broadcaster) or personalized (compiled by end user who selects all events of interest during one evening, e.g. a selection of music performances)
In order to support the innovative production workflows above, a production data model for
the ICoSOLE platform will be defined. This data model will likely be defined by extending
existing broadcast data models (e.g. the EBU Class Conceptual Data Model, CCDM), adding
the specific data functionality required for the ICoSOLE use cases. The challenges for the
data model include:
- integration of event-specific and generic production metadata
- enabling user gen metadata annotation
- keeping track of valued contributors
- support multi-platform production workflow