Skip to main content
CerebraTech AI
/work/livestock-growth-tracking
Internal testA system we're building and testing ourselves · still has an unsolved problem

Tracking livestock growth with a home-built stereo camera, instead of one fixed scale

Two CCTV cameras plus a Raspberry Pi 5, calibrated into a stereo pair as a cheaper alternative to a LiDAR camera. It photographs animals on a scale to convert into volume and weight — but there's an unsolved statistics problem: a single fixed scale only captures 100-200 animals a day out of a pen holding 5,000-10,000, which is too small a sample by statistical standards.

1600 × 900

Starting situation

Farms need daily animal weight to adjust feed formulas and make husbandry decisions, but weighing animals one at a time doesn't scale when a pen holds thousands of them.

Constraints on site

A single fixed scale lets animals step up at random, out of a pen of roughly 5,000-10,000. That only yields about 100-200 data points a day — too small a sample by statistical standards, so the resulting average isn't a reliable representation of the whole pen. This problem is still unsolved.

What we tried that did not work

A single fixed scale doesn't solve the volume problem — no matter how accurate it is at that one spot, it's still too small a sample against the whole pen. We're now planning to move to a mobile or multi-position camera setup instead (see "what we'd change next time").

What we actually did

Two CCTV cameras and a Raspberry Pi 5, custom-calibrated into a stereo pair, as a cheaper stand-in for a LiDAR camera. It's paired with the same scale — when an animal steps on, the camera captures an image and the weight together and sends both to the edge. ESP32 sensors (temperature, humidity, light, and others) send environmental data to the edge in parallel.

Results, in numbers

2 กล้อง CCTV + Raspberry Pi 5
DIY hardware standing in for a single LiDAR camera

No accuracy figures are publishable yet — both the image-to-weight accuracy and detection rate still need re-measuring once the sample-size problem is solved. The DIY hardware line is an architectural fact, not something that needs measuring.

Raspberry Pi 5 · Stereo camera (DIY calibration) · YOLOv12 · ESP32 sensors

What we would do differently next time

Next planned step (not built yet): make the camera mobile or able to relocate, covering a wider view to collect more samples instead of waiting for animals to walk onto one fixed scale — a direct fix for the sample-size problem. The image-to-weight model needs to be accurate enough first before we can remove the scale and rely on the camera alone.

Want to talk about a problem like this in your plant?

A free 30-minute call. We will tell you straight if your problem is not a good fit for AI yet.

Talk to us for 30 minutes firstSee all work

This site uses very few cookies

We use only the cookies necessary to remember your language choice and save this consent preference (your chosen theme is remembered via browser local storage, not a cookie). Our analytics tool uses no cookies and collects no personal data.

Read the cookie policy