AI-supported per-title video encoding for low-latency live streaming and dynamic video-content-workflows

Sep. 01, 2019 to Aug. 31, 2021

By 2021, the delivery of video content will account for 81% of global Internet traffic among users. In order to adapt the data volumes generated by the digital video material to the limited transmission capacities on the Internet, compression is applied during encoding prior to transmission. This compression isn't loss-free. A subjective loss-free reduction of data, required for the storage and transmission of video content, enables their transmission via today's Internet connections. Low-latency live streaming in particular places special demands on the entire transmission path and can benefit from intelligent approaches to compress content more efficiently.

The aim of the project is to research and develop new methods for AI-supported per-title encoding for video-on-demand (VOD) and low-latency live streaming. Techniques from the fields of artificial intelligence and machine learning will be used. Furthermore, these new methods will be evaluated and optimized with respect to their applicability in the economically highly relevant areas of content creation, syndication and distribution. It is predicted that the use of such methods will reduce the time required to determine the optimal settings by 90% compared to methods used today. At the same time, the average storage volumes and transmission rates could be reduced by 30% compared to classical encoding solutions, or rather the picture quality of the videos to be transmitted could be increased. From a business perspective, this will lead to significant long-term cost savings and an improvements in the quality of experience (QoE) for the end user.

fame, project, deep encode, 2019, bmbf
The DeepEncode Workflow: From the video source to an optimized video stream - via AI-supported per-title video encoding Fraunhofer FOKUS

The central research objective of the project is to research and develop intelligent and self-learning algorithms for preparation and delivery of video material,

  • which reduce the optimized adaptive video to exactly the information that the viewer can actually see on his or her end device, due to e.g. resolution or screen size,
  • which allow the optimization of streaming protocols for VOD and live streaming situations,
  • that enable the capturing of feedback loop metrics for dynamic optimization of encoding and streaming services,
  • which work format-agnostically and allow the use of existing video codecs.

Duration: 01.09.2019 - 31.08.2021

Partners: Nowtilus (GER), Nanocosmos (GER)

The project is funded by BMBF within their funding line “KMU-Innovationsoffensive Informations- und Kommunikationstechnologie” (IKT)