KNOX Vision · In operation
Video is not worth much if nobody is watching it.
Vision applies analytics to the cameras you already have installed and tells you when something that matters happens, instead of leaving you twelve screens and the hope that somebody notices the right one.
The problem
Human attention does not scale with the number of cameras.
There are old and much-quoted studies about how long a person can watch screens before they stop seeing what happens. The exact number is argued over; the direction is not. After a while, more cameras means less attention per camera.
That is why adding cameras to a site that already has plenty rarely improves the result. It improves the recording, not the watching.
What it detects
Four signals that actually change a decision.
Loitering on the perimeter
Someone staying in an area where nobody stays. It is the signal that most often precedes an attempt, and the one an operator with twelve screens never catches.
Access outside of hours
Movement in a closed area at a time when it should be empty. It does not need anyone watching: the system evaluates the rule.
Stopped vehicle
A car parked where nobody parks, or one that has been inside longer than its type of visit allows.
Attribute search
Finding “person with a dark backpack between two and four in the morning” in the recording without reviewing six hours of video. It is what turns an afternoon of investigation into ten minutes.
The indicator that matters
Cutting the noise is worth more than adding detections.
A system that sends a hundred alerts a day trains the operator to ignore them. The real work is not detecting more, it is detecting less and better: filtering out the dog, the branch, the shadow and the reflection until an alert means something.
That gets tuned on site over the first few weeks, against the rules of your operation. It does not come right out of the box, and anyone who tells you otherwise has never run a SOC.
See how the SOC runsLimits and obligations
Analytics get it wrong. That is why every alert passes through an operator before it becomes an action. A system that acts on an automatic detection by itself is a generator of false incidents.
Facial recognition and license plate reading are sensitive personal data under Mexican law. They require express consent and a privacy notice that covers them explicitly. This is not paperwork: it is the difference between a defensible operation and a fine.
Not every camera is usable for analytics. Resolution, angle and lighting determine what can be detected. The feasibility review is part of the assessment, and sometimes the conclusion is that two cameras have to be moved first.
Frequently asked
About Vision.
Does it work with our current cameras?
Do we need to change our recorder?
Does it recognize faces?
How long does it take to tune?
Next step
Try it on your own video.
We connect a couple of your cameras and show you what it detects at your site, not in a demo reel.