Sigmawave

Deployment14 min read

Adding AI to cameras you already own

A walkthrough of connecting existing recorders over RTSP, what bandwidth to expect per site, and where a rip and replace is still the honest answer.

Sigmawave deployment team
AI in the real world

Key takeaways

  • Most existing recorders already speak RTSP; the work is credentials and network routes, not hardware.
  • Run inference at the edge and send clips, not continuous streams, or the uplink becomes the project.
  • Budget a substream per camera for analysis and keep the main stream for evidence.
  • Replace cameras for optics and placement, never because the software asked you to.

The question is almost never whether the cameras are good enough. It is whether anybody still has the password to the recorder, and whether the VLAN it sits on can reach anything at all.

What you actually need from a site

  1. An RTSP URL per camera, with a substream at 640×360 or better.
  2. Credentials that are not the installer's default, and a note of who owns them.
  3. A route from the camera VLAN to one host, in one direction.
  4. A floor plan, even a photographed one, with camera positions marked.

That list takes a morning to produce and saves the fortnight that otherwise goes into discovering it one camera at a time.

Edge or centre

Run the model where the footage already is. A single edge host per site pulls the substreams locally, scores frames, and sends out only what matters, a clip, a thumbnail and a row of structured detection. The uplink then carries incidents rather than video, which is the difference between a site that works on its existing line and one that needs a new one.

When a replacement is the honest answer

Sometimes it is. A camera pointed at a wall, a lens that has been fogged since 2019, a fixed dome watching a gate it cannot see the whole of, no model recovers those. Replace for optics, angle and placement, and say so plainly rather than quietly hoping the software compensates.

Frequently asked

Does this need a cloud connection at every site?

No. The edge host keeps working without one and queues its output; what breaks is the console's live view, not the detection.

What about cameras behind an NVR that will not expose RTSP?

Most will once someone enables it. The genuine exceptions are old proprietary recorders, where the practical answer is a parallel feed rather than a fight with the vendor.

Sigmawave deployment team · 8 September 2026

Back to all articles

Our software turns real-world signals into informed decisions.

Video intelligence and synthetic data for defense, security and industrial operations. Accel connects events and evidence to coordinate responses. Terra creates the data to train and test vision AI.

Talk to our team