Your camera records 4K. You never watch it live. And when you do, it takes 8 minutes to find what you’re looking for.

That’s not a camera problem. It’s a problem of how humans process video.

( per Genetec 2024, operators spend 3 hrs/day reviewing footage manually with 2-4% success rate )

Computer Vision is the AI that processes images. In CCTV it boils down to 5 tasks:

  1. Classification — “this is a car”
  2. Detection — “there’s a car here, at these coordinates”
  3. Segmentation — “this exact outline is the car”
  4. Tracking — “this car crossed from cam 1 to cam 3 in 4 minutes”
  5. Recognition — “this car has plate ABC-123”

Modern cameras (even cheap ones) already run local detection. The YOLO model analyzes each frame in under 50ms and outputs “person / vehicle / nothing useful”. That cuts false alarms by 80-90% vs traditional motion detection.

What’s NOT Computer Vision: “AI-powered” marketing without explaining what it does. If your vendor won’t tell you which of the 5 tasks they solve and at what accuracy, they’re selling you buzzwords.

P.D. How long does your team take to find a specific event in last week’s footage? If the answer is “more than 10 minutes per search”, today’s a good day to ask what kind of AI you actually have.

📰 Source: YOLOv8 paper + Genetec State of Physical Security 2024 + Jedi vendor claims analysis.