IBC2018: Machine learning tool trialled during coverage of this year’s Royal Wedding can be used in live sports, awards and political events.
In May of this year, recognising the “real enhancement” and technical benefits of deploying machine learning (ML) based facial recognition technology, Sky set out to enhance its coverage of the Royal Wedding of Prince Harry and Meghan Markle.
“We wanted to make a statement around innovation and what we could achieve,” Hugh Westbrook, Senior Product Owner for Sky News and Sky Sports, told delegates during a session on OTT viewing experiences.
“Crucially there was exciting technology we could use together in a way that hadn’t been done before and that enabled it to come to market quickly.”
The artificial intelligence (AI) and ML techniques were deployed to identify guests as they entered the chapel ahead of the wedding.
It manifested itself in the “Royal Wedding: Who’s Who Live” function, accessible via the Sky News app or via SkyNews.com, which allowed viewers to select a guest during the live stream, identify them and get background on that person.
The service made use of Amazon Rekognition, a cloud-based image-analysis software.
Westbrook said the application “delivers tremendous value” but he recognised the importance of human curation to sense check, test and manage the comprehensive workflow.
“We know it is not going to be 100% accurate but we need that element of moderation,” Westbrook added the two phases of the workflow included testing, measurement and a creative user interface.
“Without that, we couldn’t have pulled it off.”
GrayMeta Chief Commercial Officer Josh Wiggins explained how the technology has been tried and tested with the facial recognition reaching 70% accuracy during the live stream and 80% accuracy for the VOD coverage.
The model can be marketed for live sports, awards and political events offering a richer viewing experience.
Wiggins said there is a strong business model in place with facial and object recognition. The service can be monetised based on a customisable pricing per-minute and technology selection model.
No comments yet