Scientists from the UK have presented a new system that can identify people by video. She copes with the task, even if you need to find a person from three unrelated sources.
Researchers at the University of Oxford have developed an artificial intelligence (AI) -based system that can identify people in videos. She is able to identify names and titles of people based on context and various publicly available sources, including matching audio sources with visual materials from the Internet.
The main task of AI is to identify public figures such as people who appear in television programs and films. But this principle is theoretically applicable to anyone whose face, voice, or name appears in online sources.
“Imagine watching a video and meeting a stranger there. In order to confidently identify him, you first look for clues – titles with his name and surname, other signatures. Then you search the internet for his name. The new method speeds up the search process several times, ”added the scientists.
This technique is automated and eliminates any additional manual labeling operations. Scientists have also shown that the system works well even if there are three unrelated sources to be searched.
The researchers explained that their AI will be especially useful during the exponential growth of untagged, opaque video data. Therefore, new systems are needed that can extract personal information from them without costly annotations that are filled in manually.
“The [huge] scale of the data, coupled with the lack of relevant metadata, makes indexing, analyzing and navigating this content increasingly challenging. It is no longer possible to rely on additional manual annotation, and without an effective way to navigate through the video, the knowledge bank is largely unavailable, ”the scientists added.