Tutorials · AI
How to Plan the Network for AI Cameras and Video Analytics
AI video means more streams, higher resolutions and analytics servers pulling video from many cameras. Size bandwidth, PoE and storage before you install.
Before you start
- Camera list with resolution, codec and frame rate
- Where analytics runs: camera, NVR or edge server
Step by step
Count every stream
Each camera may send a main stream to the NVR and a second stream to an analytics server or viewing client. Add them all.
Estimate bandwidth
Use the bitrate per stream (for example 4–8 Mbit/s for 4K H.265 at 15 fps) × streams × cameras, plus 30% headroom. Keep access ports at 1G and uplinks at 10G where many cameras meet.
Check the PoE budget
AI cameras with IR, heaters or PTZ draw more power. Use the maximum figure from the datasheet, and PoE+ or PoE++ ports where needed.
Place the analytics
Analytics on the camera sends only events. Analytics on an edge server (for example a NavEdge in the comms room) pulls full streams over the LAN: keep it on the camera VLAN or route between VLANs at line rate on a Layer 3 switch.
Size storage
Use your bitrate and retention days to size NVR disks; event-only recording for low-priority cameras cuts storage sharply.
Segment and protect
Keep cameras, NVRs and analytics servers on their own VLAN, with no direct Internet access.
Pro tips
- Multicast can reduce load when many clients view the same camera; plan IGMP snooping on the switches.
- Monitor uplink utilisation after go-live; analytics loads grow as features are added.
FAQ
Frequently asked questions
Do AI cameras need special switches?
No, but they need enough PoE budget per port and in total, and enough uplink capacity.
Keep learning
Related guides
CCTV bandwidth and storage calculation
Read more →Segment IoT and CCTV
Read more →Troubleshoot PoE devices
Read more →How to Use NavCloud AI Insights to Fix Problems Before Users Notice
NavCloud AI Insights turns alarms, inventory and telemetry from every site into a network health index and a short, prioritised list of findings with the action that fixes each one.
Read more →How to Set Up AI Smart Events on IP Cameras (Intrusion, Line Crossing, Human/Vehicle)
AI cameras classify people and vehicles on the camera itself, so you get far fewer false alarms from trees, rain and animals. Configure smart events and record only what matters.
Read more →How to Run AI Inference in Containers on a NavEdge Edge Server
Run AI models next to the cameras and sensors instead of in a distant cloud: lower latency, less WAN traffic and video that never leaves the site. Package the model in a container and deploy it with Docker Compose.
Read more →Stuck on a step?
Ask Navijin technical support or an authorised partner for help with your configuration.