Amazon announces Rekognition Video, which uses machine learning to analyze uploaded and real-time video streams
Frederic Lardinois / TechCrunch :
Context & Ripple Effects
This lands a week after AWS extended Rekognition itself with text detection within images and real-time face search, so the November cadence is clear: Amazon is shipping computer-vision capabilities as fast as it can productize them. Rekognition Video is the next layer up — the same recognition stack pointed at moving footage, both uploaded files and live streams.
It also closes a gap opened in March, when Google debuted its Cloud Video Intelligence API for cataloging video contents. With this launch, both major cloud vendors now sell video understanding as a metered API, and the Forbes coverage from mid-2018 shows where that leads: facial recognition tools built at incredibly low cost by anyone with a computer.
First-order effects
- Developers gain an API for analyzing uploaded and real-time video streams without building their own ML pipeline, putting live-stream analysis within reach of any AWS customer rather than only teams with in-house vision researchers.
Second-order effects
- Google's Cloud Video Intelligence API now has a direct feature-for-feature rival on pricing and latency, forcing both vendors to compete on recognition accuracy and per-minute costs for video workloads.
Third-order effects
- As the Forbes reporting on cheaply assembled facial recognition tools shows, commodity video analysis pushes the hard questions from capability to governance — who may run real-time face search on live streams becomes a policy fight, not a technical one.
The trend: Video is being converted from unsearchable footage into queryable cloud infrastructure, with AWS and Google racing to make real-time machine perception a metered utility.