Computer vision · Livestock monitoring
Eyes on the barn,even at night.
Livestock monitoring with computer vision uses cameras and AI to detect and count the animals in a barn or area and follow signals of activity and behaviour. We build systems that watch pens, feeding areas or birthing areas continuously and send the farmer flags to check, while decisions about the animals stay with the farmer and the vet.
Events
- 05:52:31Pen 2 · 8 animals counted
- 05:52:31Feeding area · empty
- 05:52:32#05 · activity below usual
- 05:58:00Flagged to the farmer
You cannot be in every barn, every hour
Many signs that something is wrong appear gradually: an animal standing apart, a feeding area visited by fewer animals, a restless female before giving birth. Checks are done at fixed times, and what happens in between, especially at night, goes unseen.
- Counts when animals enter or leave the barn are done by hand and can be wrong
- Births often start at night, when nobody is in the barn
- Changes in activity or feeding are noticed only once they become obvious
- Farm cameras, where they exist, are watched only when someone has the time
Who it is for
- Cattle, pig, sheep or poultry farms that want to keep an eye on their barns without always being there
- Breeders who want to know early when a birth or an isolated animal needs attention
- Farms with several barns or sites, where routine checks take a lot of time
- Technical teams or agricultural integrators who need a video analysis component
Step by step
How the system watches a pen, step by step
The five stages on the animation show a camera above a pen before sunrise: from the raw image to the animal flagged to the farmer.
Events
Events
- 05:52:31Pen 2 · 8 animals counted
Events
- 05:52:31Pen 2 · 8 animals counted
- 05:52:31Feeding area · empty
Events
- 05:52:31Pen 2 · 8 animals counted
- 05:52:31Feeding area · empty
- 05:52:32#05 · activity below usual
- 05:58:00Flagged to the farmer
Events
- 05:52:31Pen 2 · 8 animals counted
- 05:52:31Feeding area · empty
- 05:52:32#05 · activity below usual
- 05:58:00Flagged to the farmer
Step 1 of 5
The camera above the pen
A camera mounted high up, above the pen or the feeding area, sees the animals from above, where they hide each other the least. At night it works in infrared, so the image turns black and white and the model has to be trained on such frames too. Dust and condensation on the lens are the main enemies, which is why cleaning the cameras becomes part of the farm routine.
Which kind of monitoring suits your farm
Counting in the pen or at the gate
Counting can happen inside the pen, from frames taken at regular intervals, or at a gate, where animals crossing a virtual line are counted in and out. The first shows how many animals are there right now, while the second suits the entrance to a sheepfold or moves between barns.
Activity per group or per animal
The most robust option follows the activity of the whole pen, compared with previous days. Tracking each animal in frame gives more detail, but in dense groups the track can be lost when animals overlap, so we use it only where the stocking density allows.
Plain detection or with posture
For counting and presence at the feeder, detecting the animals is enough. For birthing pens or animals that spend a long time lying down, a model that also estimates body posture gives finer signals, but it needs more examples labelled together with the stockpeople.
What you receive at hand-over
- The detection model adapted to your species and housing, trained on daytime and infrared images and exported for the device on the farm
- The pens, feeding areas and birthing pens drawn on each camera, with the flagging rules agreed together
- Notifications set up by SMS, email or Telegram, with an image of the moment, to the people you choose
- A history by pen and day of counts, activity and flags, in which each alert can be marked as useful or not
- A measurement report from the test period, comparing counts with a human count and covering useful flags and missed situations
- Documentation, training and, typically, a plan for camera maintenance and threshold adjustment
No strings attached: tell us what the cameras should see and we'll tell you honestly whether it can be done.
Let's talkWhat the livestock monitoring system can follow
- 01
Detecting and counting animals
Counts the animals in a pen, a fold or an area and flags when the total does not match the expected herd or flock size.
- 02
Activity signals
Estimates how much the animals in an area are moving and compares it with the usual pattern for that pen at the same time of day.
- 03
Presence at feeding and drinking
Measures how long and how many animals spend at the trough or the drinker, as a signal of changes worth checking.
- 04
Watching birthing areas
Follows the birthing pens and flags changes in posture or restlessness that may indicate the start of a birth.
- 05
Flagging unusual behaviour
Can flag an animal lying down or standing apart for a long time, crowding or an animal that has got out of its pen, for checking.
- 06
Notifications to the farmer
Flags arrive by phone or email with an image from the moment, so the farmer can decide whether to go and check.
How it works
From the barn image to the farmer's check
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Human in the loop
- Result
- 01
Input
Cameras in the barn
The cameras covering pens, folds or feeding areas send images continuously.
- 02
AI
Detection and posture
The model finds the animals, counts them and estimates their position and movement.
- 03
Action
Comparison with pattern
Activity is compared with the pen's usual pattern, and deviations become flags.
- 04
Human in the loop
Farmer's check
The farmer looks at the image and decides whether to go to the animal or call the vet.
- 05
Result
Log and statistics
Flags, counts and activity are kept in a history by pen and day.
What data livestock monitoring needs
- Real images from your barns, at different times, including at night, with the usual animals and stocking density
- Labelled examples of the situations of interest, for example an animal lying down, feeding or in labour, marked together with the people who look after the animals
- Information about the herd or flock, pens, feeding schedule and farm routine, so flags take them into account
- Cameras suited to the environment: resistant to dust and humidity, with infrared for night and mounted above the areas of interest
- Access to camera streams (for example via RTSP or ONVIF, where the equipment supports them) and a stable connection on the farm
- An edge device or local server, so the analysis keeps working when the internet connection is weak
Possible integrations
- IP cameras with an RTSP stream and infrared illumination
- ONVIF-compatible cameras
- NVRs and VMS systems
- NVIDIA Jetson edge devices
- Notifications by SMS, email or Telegram
- Farm management applications, via API, where the vendor offers one
- Feeding or climate control systems, as a source of data on schedules and conditions
- MQTT
- PostgreSQL databases
- IP cameras with an RTSP stream and infrared illumination
- ONVIF-compatible cameras
- NVRs and VMS systems
- NVIDIA Jetson edge devices
- Notifications by SMS, email or Telegram
- Farm management applications, via API, where the vendor offers one
- Feeding or climate control systems, as a source of data on schedules and conditions
- MQTT
- PostgreSQL databases
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays with people
- Any decision about an animal's health, treatment or moving it, which belongs to the farmer and the vet
- Checking flags on the spot, because an image does not show everything a person sees in the barn
- Defining what unusual behaviour means for each species, age group and barn
- Adjusting the flag thresholds so that alerts are useful and not too frequent
Limits and operating conditions
- The system does not diagnose: it flags visible changes, and their cause is established by the people who know the animals
- In dense groups, animals that overlap or hide one another are harder to count
- Recognising a particular animal within the group is not guaranteed; without visible ear tags or other markings, the system works mainly at pen or group level
- Dust, condensation, dirt on the lenses and poor light can reduce image quality, so camera maintenance is part of the project
How we evaluate livestock monitoring quality
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Counting accuracy
- The number given by the system compared with a count made by a person on the same images, across different pens and times
- Useful flags
- The share of flags the farmer confirms as situations worth checking, recorded for each alert
- Missed situations
- Events noticed by staff, for example a birth, for which the system sent no flag, gathered in a shared log
- Day and night stability
- Results compared separately on daytime images and infrared images, across different barns
How we implement livestock monitoring, in 4 steps
- 01
We choose what to follow
We agree with the farmer which situations matter, in which barns and who receives the flags.
- 02
We install or check the cameras
We review the existing cameras or propose new positions and collect real images, by day and by night.
- 03
We train and test in parallel
We adapt the models to your animals and barns and let them run alongside the usual checks, comparing flags with what staff observe.
- 04
We switch on notifications and monitor
We activate notifications, adjust thresholds based on the farmer's feedback and retrain when the barns or the herd change.
Use cases
The scenarios below are conceptual examples, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Dairy farm
Calving pens overnight
Infrared cameras watch the calving pens, and when a cow shows prolonged restlessness the stockperson on duty receives a notification with the image, to decide whether to go to her.
Conceptual example - Pig farm
Activity at the troughs
For each pen, the time the animals spend feeding is tracked, and drops compared with previous days appear in a morning report checked by the farmer.
Conceptual example - Sheep farm
Counting at the fold gate
A camera above the gate counts the sheep coming in for the evening, and any difference from the flock size is sent to the shepherd, who checks whether an animal is missing.
Conceptual example
Useful guides
Before you decide, read
Frequently asked questions
What is AI livestock monitoring?
It is the use of cameras and computer vision models to detect and count animals and to follow signals of activity and behaviour in barns, at feeding or in birthing areas. The system sends flags to check, and the farmer and the vet decide what happens next.
Can the system tell that an animal is ill?
No. It can flag visible changes, for example an animal standing apart or eating less than usual, but it does not diagnose. Interpretation and any treatment remain with the farmer and the vet.
Which species does it work for?
The principle can be applied to cattle, pigs, sheep, goats or poultry, but the models have to be adapted to each species and each type of barn. We start from images from your farm and check what can realistically be observed in the conditions there.
Can it recognise each individual animal?
We do not promise that. In most situations the system works at pen or group level, because animals look alike and hide one another. Where ear tags or other markings are clearly visible, we can check together whether identifying individual animals is feasible.
What about the people who appear in the images?
The system is designed for animals and does not identify the people who work on the farm. Cameras are pointed at pens and feeding areas, people can be blurred in stored images, and employees must be informed about the cameras, in line with the obligations on video surveillance at work.
Does it work without good internet on the farm?
The analysis can run on a local device on the farm, so it does not depend on a fast connection. A modest link, for example over the mobile network, is enough for notifications, and images do not need to be sent to the cloud.
Can’t find the answer? Write to us and we’ll reply directly.
Want to know what happens in the barn when you are not there?
Tell us which animals you keep, how many barns you have and which situations you want to follow. We will tell you what can be observed with cameras and what would be needed.
Related solutions
- People and object countingAnonymous counting of crossings, zone occupancy or the items in a batch, without identifying anyone.
- Video event detectionVisual rules that flag entries into restricted zones, missing protective equipment or fallen objects, for human review.
- Edge AI for computer visionVideo analysis on devices installed on site: fast response, less network load and images that never leave the building.