Technology
The site is the data plane
One rule shapes the whole system: video never leaves the site unless somebody is watching it or an alert fires. Everything needed to detect, record and deter runs on a single sealed unit in a cabinet on the farm. The cloud holds identity, history and the dashboard — nothing the perimeter depends on to work.
The intelligence
Powered by deep learning. Built for the wild.
Six things stand between an animal stepping into frame and a farmer losing a season. All of them happen on the unit at the site, in under a second, whether or not the internet is up.
- 01
Trained neural detection
Every camera feed is processed frame by frame by a custom-trained deep learning model — purpose-built to recognise wildlife, not just to detect motion. It knows the difference between a swaying branch and an elephant.
- 02
Real-time inference, on-site
The model runs directly at the site, analysing live video as it happens. No cloud upload, no processing delay — the moment an animal enters frame, the network is already identifying it.
- 03
Species-level recognition
It does not just report that something moved — it classifies what moved: elephant, wild boar, leopard, deer, person. Each class is mapped to its own response profile, because what turns an elephant back is not what you want firing at a person.
- 04
Strobe-triggered deterrence
The instant a detection is confirmed, the unit fires a strobe-light and audio sequence tuned to that species — engineered to disrupt and redirect the animal before it reaches the crop or the settlement.
- 05
Immediate alerts to you
The same instant the deterrent fires, an alert reaches your phone with a snapshot and a short clip as proof. You see what the system saw, as it happens.
- 06
Night vision, full accuracy
The model is trained across day and night conditions, so accuracy holds in total darkness — which is when most of this actually happens. At the pilot site, most intrusions began after 8 PM.
The model architecture, the training set and the thresholds behind it are the product of a season of field data, and are proprietary to Vision Logix. We are glad to go deeper under a mutual NDA.
Reliability
What happens when the internet dies
Rural sites lose connectivity. The system is built on the assumption that they will.
Detection keeps running
Nothing in the detection path makes a network call. Watching, recognising, recording and the alarm are all local to the site.
The deterrent still fires
There is no round trip to a server between the animal appearing and the sound playing. The unit that sees it is the unit that responds.
Alerts queue, they do not vanish
Every event is written to durable local storage first and sent afterwards. Six hours offline means six hours of alerts back-filled the moment the link returns.
It boots without a network
The last known-good configuration is held on the device, so a site that reboots during an outage comes back up watching.
Security and privacy
The cameras never touch the internet
Isolated camera network
Cameras sit on their own network segment, reachable only by the on-site unit. There is no camera of yours exposed to the public internet for anyone to find.
Outbound only
The site reaches out over an encrypted private link. Nothing is left listening at your router — no port forwarding, no static IP, no open door.
No standing credentials on site
The unit in the field holds a revocable token scoped to that site alone. A stolen box is a stolen box, not a way into anything else.
Footage stays on site
Only alert media — the annotated snapshot and a short clip — leaves by default. The archive stays on your ground and is fetched on demand over the private link.
Deployment
What actually goes on the ground
Camera count, placement and deterrent siting are set per site during survey — a ten-acre holding with one approach path is a different install from an estate boundary.
- On-site unit
- One sealed, actively cooled compute cabinet
- Power
- Low-draw, viable on solar with battery backup
- Cameras
- Up to 8 per unit, day/night with infrared
- Coverage
- Regions defined per camera, typically to ~80 m
- Deterrent
- Relay-driven audio and high-intensity light
- Storage
- On-site drive, retention managed automatically
- Connectivity
- Any ordinary broadband or mobile link, outbound only
- Ongoing
- Remote configuration, health monitoring and model updates
The control plane
Small on purpose
The cloud side does no video processing and holds no camera credentials. It exists so a person can log in, see history and watch a live feed — and so a site's settings can be changed without driving to it.
Identity and access
Accounts, roles and strict per-client separation.
Detection history
Every confirmed event, with the frame that triggered it.
Alert delivery
Notifications to whoever the site is configured to reach, on their phone.
Configuration
Cameras, regions, sensitivity and night hours edited centrally and picked up by the site within a minute.
Fleet health
Continuous status from every deployed unit — cameras up, storage, temperature, queue depth.
Live view
Any camera on demand, opened per session and closed after.
Want to see it working?
We would rather walk you through footage from a live site than talk in the abstract. We are glad to go deeper on the engineering under a mutual NDA.