AI Camera System 2027 Roadmap

The recording server's GPU misconfiguration (the "ticking time bomb") has been corrected, and IntuVision has been upgraded to version 18. We have established a direct relationship with IntuVision. We will use Dave's remaining prepaid hours to learn his bespoke access-control work.

Effort · SHardware · n/a

IntuVision offers a redundant/failover-node add-on. The server's 100% CPU has now been traced to misconfiguration and corrected, rather than hardware nearing failure, so the failover box is deferred for now. We can revisit it if stable data later points to a hardware limit; it is a light lift on our side, with most of the work handled by IntuVision support.

Effort · SHardware · M

Today, IntuVision's detections (after-hours motion, tailgating, tamper) log quietly to a screen nobody watches. Plan: route high-priority detections through an open-source home-automation hub to trigger something the desk can't miss: a physical alert (siren, flashing light), a push/Telegram/text notification, or a mix. Building on the open-source suite lets us tune it to exactly what the desk needs, and the same alerting layer is reused by (D). It also gives us a common framework to run (D) head to head against IntuVision on the same detections. Firm scope follows once we finalize what the desk needs to see and hear.

Effort · LHardware · S

Open-source detection engine, with no per-camera license fee (unlike IntuVision's per-camera annual licensing). Not a replacement for IntuVision: no tailgating/anti-passback or line-crossing, so IntuVision stays on access-controlled areas. The proof of concept, on the pet area/dog park catching off-hours use, is done and was successful, carried out free of cost while I came up to speed on the platform. It confirmed Frigate can do useful detection on our cameras. Taking it to production is item E.

Effort · SHardware · n/a

Take the proven proof of concept to production. A dedicated refurbished PC sits downstairs on the camera network running Frigate, the deployment is rolled out to about three additional cameras, and its detections are wired into the item C alerting pipeline so the front desk actually receives them. Running through the item C framework lets us compare Frigate head to head against IntuVision on the same cameras and detections, giving the board direct evidence of where the open-source engine matches the licensed platform and where it does not. If we ever stand the deployment down, that same box can be repurposed as the failover node in item B, so the hardware is not wasted either way.

Effort · LHardware · S

A second-pass layer that ties into item C. When item C flags a high-priority detection, a vision AI model reviews the same footage to add context, filter false positives, and describe in plain language what it sees for the front desk. It builds on the item C alerting pipeline rather than standing alone, so it is a natural follow-on once C is in place. Local inference on a small GPU box keeps footage on the camera network; a hosted vision API is a lower-hardware alternative but sends footage off-site.

Effort · LHardware · M