Amazon Web Services (AWS) has unveiled a new video content recognition offering, Amazon Rekognition Video, which uses deep learning to identify objects and people either in stored video or in a live stream. The new product was announced at Amazon’s AWS re:Invent event in Las Vegas alongside a host of other cloud-based AI-driven products, and will be another option for publishers looking into video content recognition for content filtering and brand safety assurance.
Amazon Rekognition was launched last year, initially only handling image recognition and analysis, though it did offer interaction with live feeds by analysing still images taken from those feeds. Now this services has been expanded to cover video as well, both live streams and video stored on Amazon’s S3 cloud storage. According to the announcement from AWS, “With Rekognition video, you can accurately detect, track, recognize, extract, and moderate thousands of objects, faces, and content from a video.”
Amazon say that Rekognition is unique as it uses “the complete context of visual, temporal and motion of the video” to infer what is appearing on screen. For example, the service feature can identify that there is a man, a car, and a tree in the video, as well as deduce that the man in the video was running to the car.
The software could prove useful for publishers concerned about brand safety, specifically those who receive large amounts of user-generated content (UGC) which are difficult to manually review before being posted online. Amazon also say Rekognition allows organisations managing user-generated content, such as social media or dating apps, to automatically detect explicit or suggestive content in videos and create their own rules around what is appropriate for the culture and demographics of their users. Users can choose to opt in to content moderation, which flags up videos that violate set criteria and passes them on for manual review.
The brand safety possibilities look like they could be a solution for the type of brand safety problems being encountered by YouTube, although the company claims to have its own video recognition software. However, other publishers and live streaming websites might be tempted by Amazon’s offering, both for categorising/segmenting inventory and for brand safety.
Video content recognition software is certainly nothing new, but the technology is still far from perfect. For example, researchers earlier this year found that Google’s “cloud video intelligence” tool could be fooled by inserting single frames of unrelated content into a video. If Amazon’s product manages to overcome these hurdles it could be a more effective tool for publishers to ensure user-uploaded and live streamed content remains within their standards, allowing them to offer a more brand safe environment for advertisers.