Livestock count
Farmers want to know the flock is all there without walking the field twice a day; a missing animal is often only noticed at the next full count.
Timed capture. Generative AI counts 24 sheep and notes a person with a dog walking the flock. Count unchanged since morning; logged.
A timed capture. No PIR or radar involved; every frame is uploaded and read by generative AI.
On a timer the camera captures the field and every frame is uploaded. Generative AI counts the sheep in frame and notes any people, dogs or vehicles, returning a number rather than a verdict. The count is charted, and a rule on the count, a fall against the running average or below a floor you set, raises the alert.
- Wake
- Timer only; PIR and radar off
- Schedule
- Capture every 4 hours in daylight, once at night with IR. Every frame is uploaded.
- On the camera
- None. Counting needs every frame, so nothing is filtered on the camera.
- Your criterion
Count the sheep visible in the field. Note any people, dogs or vehicles.
Written in plain English. The model reads every escalated frame against it.- Alert rule
count.sheep < rolling_average − 3 OR count.sheep < 20 · hold-off 4 h- Who is told
- Push to the farmer with the count and the image; weekly count summary by email.
- Evidence kept
- Every count kept as a time series; frames kept for 90 days.
MachairanowCount is 17, down from 24 at 14:00. A livestock trailer is at the gate.
A readable account of what triggered it, with the frame attached. No feed to watch, no captures to trawl through.
Jeremy Clarkson: 11 sheep stolen from Diddly Squat, discovered only at a flock count (September 2026)