Numbers
With on-camera video analytics, the image is analysed in the camera’s memory and discarded. Only counts are stored, and they leave the camera only when you fetch them. This is ACAP edge processing as Vicodis uses it.
On-camera counting compared to server-based video analytics
Three architectures side by side: analytics on the camera, on a local server and in the cloud. What leaves the device in each case, what it costs to run and what happens when the network fails.
What travels over the network
On the cameraedge analyticscounts only
Local serverserver based video analyticsvideo stream
Cloudvendor platformdata off site
Overview
Three architectures, side by side
With edge analytics, people counting moves the analysis to where the image is captured. The alternatives move the image to where the analysis runs.
| On the camera | Local server | Cloud | |
|---|---|---|---|
| Where the image is analysed | Inside the camera | On a server in your building | In the vendor’s data centre or on a sensor that reports there |
| What crosses the network | Finished counts | Continuous video streams | Counts or images, depending on the product |
| Extra hardware | None | Server, storage, licences | Often a gateway or dedicated sensor |
| Internet needed | No | No | Yes |
| Who runs it | The camera | Your IT | The vendor |
Data
What leaves the device in each case
Video
CCTV based people counting sends the full video stream to the server. The image leaves the camera, travels through the network and is processed — and often stored — somewhere else.
Data off site
Depending on the product, counts, metadata or images leave the building and are processed by a third party, under its contract terms and in its data centres. See cloud people counters compared.
What this means for data protection is set out on people counting and the GDPR. Dedicated sensors without a camera image are compared on 3D stereo people counters.
Infrastructure
Network load and infrastructure cost
A video stream for analytics typically needs several megabits per second, around the clock, for every camera — plus a server that can decode and analyse all of them at once. A count needs a few bytes when someone crosses the line.
On-camera counting therefore needs no extra server, no additional storage and no reinforced network. The processing power is already in the camera you bought.
Per camera, around the clock
Video stream to a serverseveral Mbit/s
Counts from the cameraa few bytes per crossing
Server hardwarenone needed
Failure
What happens when the connection drops
If the counting happens elsewhere, a network failure means a gap in the figures — unless the product buffers locally. If the counting happens in the camera, the connection is only needed to read the results.
- On the camera: counting continues and is stored locally; nothing is lost.
- Local server: no stream, no count — the gap stays.
- Cloud: depends on the product; many buffer for a while, some do not.
Operation
Maintenance and updates
A server needs an operating system, security patches, backups and eventually replacement. A cloud service is maintained by the vendor, along with the subscription. An app on the camera is updated like the camera’s firmware: a package per device, which tools such as AXIS Device Manager can roll out to many cameras at once.
Vicodis is an app of this kind — see the Vicodis Visitor Counter, and for sites without any internet, people counting without cloud, server or internet.
- Camera app: one package per camera, data and licence stay.
- Server: operating system, patches, backups, hardware cycle.
- Cloud: handled by the vendor, tied to the subscription.
FAQ
Questions about the architectures
01
What is edge analytics in people counting?
Edge analytics means the video is analysed where it is captured, inside the camera, instead of on a server or in the cloud. For people counting, the camera’s own processor detects people and counts line crossings; only the resulting figures leave the device. Vicodis uses this approach as an ACAP application on Axis cameras.
02
Is server-based video analytics ever the better choice?
Yes, when you need analysis that a camera cannot do on its own, for example correlating many cameras or running heavy models on older cameras without a deep-learning unit. The cost is a continuous video stream to the server, a machine to operate and maintain, and image data leaving the camera.
03
Does on-camera counting need a special camera?
It needs a camera with a deep-learning processing unit that runs AXIS Object Analytics, which covers Axis cameras on the ARTPEC-9, ARTPEC-8 and CV25 platforms. Many cameras installed in recent years already qualify. Older models without such a unit cannot run the analysis on the camera and would need replacing.
How many visitors does your entrance really have?
Tell us your camera model or your project — we check compatibility and come back with a concrete proposal.