Resolution
A 4K stream costs roughly four times the AI capacity of a 1080p stream. Moving analytics to a lower-resolution sub-stream is the single biggest saving available.
How many AI channels do you actually need?
Camera count alone does not size an AI box - resolution, analytics frame rate, algorithms per stream and retention all consume capacity. Move the sliders and see the model class and storage you should be planning for.
Live estimateBased on 40+ algorithmsNo email required
Tell us about the cameras
Everything is estimated in your browser. Change any value and the result updates instantly.
cameras
Only count cameras that need analytics.sites
Each site usually needs its own AI box.Higher resolution costs more AI capacity and more storage.
Analytics rarely needs full frame rate - people and vehicle events are reliably detected well below 25 fps.
per camera
How many detections must run on the same stream at once.days
How long footage must stay on the box.Your estimate
—
Planning estimate only. Assumptions: H.264 continuous recording, usable capacity after filesystem overhead, and an AI load model of a 1080p stream at 25 fps running one algorithm. Real figures depend on scene motion, codec and retention policy.
Sizing
The four things that change your channel count
Most sizing mistakes come from underestimating one of these.
Analytics frame rate
Dropping from 25 fps to 12 fps typically halves the load with almost no loss for people and vehicle events. Only fast-motion detection such as plate reading needs more.
Algorithms per camera
Running helmet, intrusion and smoking detection on one stream costs more than running one of them. Share a stream only where it genuinely makes sense.
Retention and sites
Storage scales linearly with days, and every site usually wants its own box so that recording survives a network outage.
Where this estimate stops being enough
A calculator can size capacity. It cannot tell you whether your chosen algorithm will actually work on your camera angle. That is why every enquiry gets a review from an engineer who checks field of view, lighting, mounting height and network before recommending a bill of materials. Send us the camera list and we will tell you which channels are realistic and which will need repositioning.
FAQ
Frequently asked questions
Is the recommended model binding?
No. It is the model class to plan against. Final selection also considers alarm I/O, storage bays, PoE budget and whether you need redundant power.
Why does an AI box have fewer AI channels than camera channels?
A box may accept 32 camera streams but only have the neural compute to analyse a subset at full frame rate. XCAN boxes let you assign algorithms per channel so you spend capacity only where it matters.
Can I add analytics to cameras in different buildings?
Yes. Cameras stream to the box over the network, so they can be on any reachable VLAN. Bandwidth and latency between sites is usually the limiting factor - use one box per building and a central management platform.
What if I need more channels later?
The XCAN platform is licensed per channel and boxes can be stacked. Start with the model that covers today plus a realistic 12-month growth step rather than buying for a worst case that never arrives.
Want a real bill of materials?
Send the estimate with your camera list and site plan. We will return a channel plan, a hardware list and a realistic delivery date.