Definition
Edge AI means performing analysis on a device located where the data is produced, instead of sending it to a remote data centre. In vision systems the video is analysed on site, and at most event information leaves the premises.
Why it matters for video
Streams from a dozen or several dozen cameras represent traffic that usually cannot sensibly be pushed off site, neither in bandwidth nor in cost. On-site processing solves that at the source.
The second reason is response time. An alert about a person entering a machine's working area is useful within seconds, not after buffering, transmission and processing at a remote facility.
Data and independence
With local processing, footage never leaves the site, which simplifies internal agreements and the data protection impact assessment. For some customers this is a precondition for the entire project.
The system also works with the link down. Connectivity then serves remote support and reporting rather than detection itself, so an internet outage does not switch the safeguard off.
When the cloud wins
The cloud wins for rare and computationally heavy tasks: training models, historical analysis over large datasets, aggregating data from many sites.
A sensible split is therefore mixed: detection and response locally, training and aggregate analysis off site, on data stripped of imagery.
Frequently asked questions
Does edge AI require powerful hardware on site?
It requires a device suited to video analysis, sized to the number of cameras and concurrent scenarios. Scale follows the number of streams, not the size of the plant.
Is local processing slower than the cloud?
The opposite. It removes the transmission time, which is significant for video. The limit is the device's capacity, not the connection.
What happens when the internet fails?
Detection and signalling continue. External reporting and remote viewing pause and resume once the link is back.