GrizEyes is the vision sense of GrizCam.
Strapped to a tree or banded to a fence post, one GrizCam watches a wide field around it, not a narrow cone. It does not wake every time the grass sways or a cloud’s shadow slides across the trees. An always-on, low-power watch system we call ANA triggers on how something moves rather than on a passing warm patch, and it learns its own scene from the moment it is deployed, with no zones to draw. Only when ANA decides the motion is worth a look do the 4K cameras and the on-device AI wake to capture and label it, so the battery is spent on animals, people and vehicles, and far fewer empty frames reach you.
How ANA decidesEach ANA event goes out over cellular, satellite or a private-area mesh network to the GrizCam Portal, where you review, search and get alerted on activity across a whole deployment as it happens, from any browser.
For deeper work, GrizEyes in GrizCam Desktop lays out every capture beside the audio from the minutes before and after it, sound an ordinary camera trap never records, so you hear what led up to the moment and what followed. It also opens photos and video from other camera traps, such as Browning and Reconyx.
Strap it to a tree, band it to a fence post, bolt it to a gate frame, a corner post, a pipe stand, a shipping container or the side of a building — a trunk in the timber and a pole on a well pad are the same job to the camera. Wherever it goes, it covers every approach around it, from any angle and at any bearing. The only thing it cannot see through is whatever it is mounted to, and even there ANA stays aware of movement.
Nearly every camera trap ever sold decides when to record the same way: a passive infrared sensor watching for a change in the heat pattern crossing a segmented lens. It is cheap and it draws almost nothing, which is why it won — a dumb switch that can sit on a battery in the woods for a season. But it never sees an animal. It notices that a pattern of warmth moved, and everything it gets wrong follows from that.
ANA runs three different types of very low power sensor, each strongest where the others are weakest, and fuses them into a single question: is this motion alive, or not alive? None of them depends on an animal being warmer than its background, so the failures that define a heat trigger are simply not failures here.
It fires on the signature of the motion — purposeful travel rather than vegetation moving in place. That is why an unfamiliar subject still gets recorded: nothing has to recognize it first. And behind the device, where the lenses do not reach, ANA is still aware.
The picture itself comes from two 4K low-light imagers whose fields overlap and are stitched into one continuous scene — so what gets captured is a full-resolution look at whatever earned it, not a thumbnail of a heat blob. Because the two are mounted a short baseline apart, the only place they can disagree about where something sits is very close in; that disagreement is an angle that falls away as 1/distance, which is why the seam is invisible out in the scene.
They are chosen for the hours that matter. Most of what moves in a landscape moves at dawn, at dusk and after dark, and that is precisely when a camera built for daylight gives up — so these hold usable color far down into low light rather than dropping straight to a gray infrared frame. And detection close to the device does not depend on visible light at all, so the decision to record is never waiting on the scene to brighten.
Every one of these failures is silent. A false trigger costs you a wasted frame and you can see it in the folder. A miss leaves nothing behind at all — no photograph, no timestamp, no record that anything was ever there. You cannot audit the animals your camera never told you about, and a season of data can be shaped by a gap nobody can measure.
Bolting a classifier onto a heat trigger does not fix this. It moves the problem: the model only ever votes on what the trigger already let through, and its accuracy falls from roughly 96% at the locations it was trained on to 69% at new ones — failing hardest on exactly the small, distant, half-hidden subjects a heat trigger was already losing.
An eye does not send the brain a picture. It sends a small set of pre-computed signals, and the brain engages fully only when one of them fires. GrizCam is built the same way: something inexpensive stays awake and decides what is worth the cost of a full look. That division is what makes long unattended deployments practical.
A camera that must identify a subject before recording misses exactly what matters most — the half-hidden animal, the one facing away, the thing its training never saw. GrizCam captures on the signature of the motion: purposeful travel rather than vegetation moving in place. An unfamiliar subject is still recorded, and then becomes an example you can train on.
From the moment it is deployed it builds a picture of which parts of its view are trustworthy and which are habitually noisy — the branch that moves in every breeze — and it grows more selective the longer it watches. No zones to draw. No setup.
Detection close to the device does not depend on visible light. What happens near the camera at night is still noticed.