NEWS / ARTICLE

Home / News / Article

AI video analytics on the camera or in the cloud: choosing between edge and centre

324 wordsKnow-how

Beyond the algorithm itself, one question is unavoidable in video analytics: where should the processing actually happen? The location changes bandwidth, latency, privacy and cost. This is a comparison of the usual deployment options and their trade-offs — a guide to the thinking behind the choice.

Four places, from front end to cloud

Roughly, there are four layers. Front-end analytics run inside the camera and return only results. Edge computing places a device on site that aggregates several streams before analysing them. A central server processes everything in the equipment room. The cloud takes the data or the results off site. The closer to the front end, the less bandwidth and the lower the latency; the closer to the centre or cloud, the more concentrated the compute and the easier unified management and algorithm updates become.

What the trade-off turns on

  • Bandwidth: sending all raw video to a centre or the cloud puts heavy load on links when many streams combine; producing results at the front end cuts that sharply.
  • Latency and resilience: scenarios needing instant response lean on front end or edge, and keep working if the WAN drops.
  • Privacy and compliance: where facial or other sensitive data is processed, and whether it may leave the site, is a premise to settle first.
  • Compute and cost: front-end compute is limited and upgrading means replacing hardware; central deployment scales and updates more easily but needs room and link investment.

Reality is usually a mix

Few projects use just one. A common pattern is real-time, lightweight decisions at the front end or edge, and aggregation, statistics and model iteration at the centre or in the cloud, each layer doing its part. We prefer to separate what must be instant, what can wait, and what data cannot leave the site, then decide what runs where — an architecture driven by requirements is steadier than pushing everything to the cloud or cramming it all into a camera.

← Back to news Have a similar need? Talk to us