Automated Mobility Scoring for Dairy Cows: How the Camera Systems Compare

A trained scorer standing at the parlour exit sees your herd perhaps once a month, and sometimes only once a quarter. A camera mounted above the same exit race sees every cow at every milking. That difference is the whole case for automated mobility scoring on dairy cows. Whether it pays depends on your herd size, your parlour layout, what your milk buyer accepts, and how quickly someone on the farm acts on the alerts.

This page compares the main camera approaches available to UK dairy farms, works through the costs for a typical herd, and flags the things salespeople tend to skip. If you want the wider picture first (collars, boluses, pedometers, and where cameras sit among them), start with our overview of AI livestock health monitoring.

What the camera is actually doing

Every system on the market follows the same broad sequence. A camera watches cows walking in single file, usually along the exit race after milking. Software picks out each animal, ties it to an ID (from the parlour’s EID reader, or in some cases from coat pattern), tracks how the back, head and legs move across several strides, and gives that cow a score.

Most UK systems report on the AHDB Mobility Score scale so the output matches what a RoMS-registered scorer would write down:

The value is not that a camera scores better than an experienced person on a single day. It is that it scores every day, so a cow drifting from 1 to 2 gets flagged within days rather than at the next monthly session, which is when early treatment of sole ulcers and white line lesions has the best results.

2D versus 3D: the main technical split

The systems fall into two families, and the difference affects installation, accuracy and price.

2D (standard video) systems. CattleEye, developed in Belfast, is the best-known UK example. It uses a single overhead camera, often a standard security camera, looking down on the exit race. The AI is trained on very large amounts of footage to read back arch, head bob and stride from an ordinary video image. Hardware is cheap and simple to fit, and much of the work happens in the cloud.

3D (depth-sensing) systems. HerdVision, developed in the UK, uses a 3D camera that measures the shape of the cow as well as her movement. That depth data lets the same unit produce body condition scores alongside mobility, which is a real advantage if you currently pay for BCS separately or don’t do it at all. The camera is more specialised, and positioning matters more.

There is also a trap to avoid here. Some 3D cameras from the big parlour manufacturers, the DeLaval BCS camera for example, measure body condition only, not mobility. If a rep tells you a camera “monitors cow health”, ask directly whether it outputs a mobility score per cow per milking.

Feature lists change quickly in this market, so treat this table as a guide to the types of system, not a fixed spec sheet for each brand. Ask each supplier for their current one.

What your parlour needs before any of this works

The camera is the easy part. Whether your exit race suits it is the hard part.

Cows need to walk in single file, at a natural pace, on a reasonably flat surface, for at least two or three body lengths in view of the camera. A race that bends sharply, a step down out of the parlour, a gate that bunches cows together, or a dog chasing them out will all degrade the scores. If cows run out of a rotary or trot out of a herringbone in pairs, expect the supplier to suggest changes.

You also need:

• Reliable EID reading at or near the exit, so each score is attached to the right cow. A misread rate of even a few percent means alerts for the wrong animal.
• Power and internet at the camera position. A 4G router works on many farms, but test the signal in the parlour, not in the yard.
• Sensible lighting. Strong low sun shining straight down the race at evening milking can cause problems for some systems, so mention it at the survey.
• Robotic herds: automated scoring is possible, but the camera has to sit where cows leave the box or walk through a selection gate. Ask for references from other robot farms specifically.

Worked example: a 220-cow herd in Shropshire

Here are the numbers for a plausible case. Swap in your own herd size and quotes.

The herd: 220 cows, milked twice a day through a herringbone, yielding 9,000 litres. Currently scored by a RoMS scorer once a month. The last four sessions showed 28 to 32% of cows at score 2 or 3, which is roughly the UK average found in national surveys.

The cost of lameness: the figure most often quoted in the UK is around £330 per case once lost milk, extra culling, reduced fertility, treatment and labour are added up. It varies a lot by lesion type and how late the case is caught, so use it as a working figure, not gospel.

The assumption that matters: the camera doesn’t stop cows going lame. What it can do is get them treated earlier, so cases are shorter and fewer turn chronic. Say earlier detection and treatment cuts the average cost per case by a quarter, and the herd’s annual case count falls by 15% because fewer cows relapse.

The illustrative system cost: suppliers mostly price per cow per month or per year, sometimes with an installation fee. For this example we’ve assumed £2.50 per cow per month plus £1,500 installation. That is not a quote from any named supplier. Get real quotes and replace it.

Annual lameness cases (before)       220 cows x 0.55 cases/cow/yr  = 121 cases
Cost of lameness (before)            121 x £330                    = £39,930

Cases after 15% reduction            121 x 0.85                    = 103 cases
Cost per case after 25% reduction    £330 x 0.75                   = £247.50
Cost of lameness (after)             103 x £247.50                 = £25,493

Gross saving                                                         £14,437

System subscription                  220 x £2.50 x 12              = £6,600
Installation (spread over 3 years)                                   £500
Monthly RoMS scoring dropped?        Assume kept quarterly instead, saving £1,200

Net annual benefit                   £14,437 - £6,600 - £500 + £1,200 = £8,537

At these assumptions the system pays for itself comfortably. Now halve the improvement (a 7.5% case reduction and a 12.5% cost-per-case reduction) and the gross saving falls to roughly £7,700. That only just covers the subscription. The lesson is plain: the return depends almost entirely on whether someone acts on the alerts within a few days. A camera producing a daily list nobody reads costs £6,600 a year and saves nothing.

Daily use: what the output looks like in practice

Most systems give you a phone app or web dashboard and some form of daily alert. The useful pattern is a short list of cows whose score has worsened since their recent baseline, rather than a list of every lame cow in the herd. Seeing 60 cows flagged every morning teaches people to ignore the list. Seeing six cows that have just moved from 1 to 2 gets them into the crush.

Set up a routine before the system goes live:

1. One named person checks the alert list after morning milking.
2. Flagged cows get drafted (automatic shedding gates make this far easier) and lifted that day or the next.
3. The foot trimmer or herdsperson records the lesion found, ideally in your herd software, so you can see whether alerts are turning into real findings.
4. After a month, compare the camera’s flags against what was found in the crush. A good system should show a clear majority of score-2 alerts turning up a lesion.

That fourth step is how you find out whether the system works on your farm, which matters more than any trial result in a brochure.

Does it replace your mobility scorer?

Not completely, and not straight away. Red Tractor’s dairy standard expects regular mobility scoring by a trained person, and many milk contracts (Arla, Müller and the retailer-aligned pools each have their own wording) set their own scoring requirements. Some buyers now accept automated scores as evidence. Others want a human scorer at set intervals regardless.

Phone your milk buyer’s technical team before you sign anything. Ask whether they accept automated mobility data for their scheme, from which named systems, and whether you still need a RoMS scorer and how often. The answer could change the worked example above by a four-figure sum either way.

A sensible middle path many farms take is to keep a human scorer quarterly as a calibration check. Score the same day the camera does, compare the two, and raise any gap with the supplier.

Cameras versus leg sensors

It’s worth ruling sensors in or out. Leg-mounted accelerometers such as IceRobotics CowAlert, or the lameness features in some collar systems, detect changes in lying time and step pattern. They work in any housing and don’t depend on the race layout. The catch is that you’re buying, fitting and replacing a device on every cow.

Cameras win when you have a good exit race and want one piece of kit covering the whole herd. Sensors win when the parlour layout defeats cameras, when you already run collars for heat detection and can add lameness as a module, or when cows spend long periods outdoors and don’t pass a camera daily. Spring-calving grazing herds walking long tracks to the parlour still pass the exit race twice a day, so cameras can suit them well, provided the race doesn’t turn into mud.

Questions to put to every supplier

Ask each company the same questions in writing and compare the answers side by side:

• Which mobility scale do you report on, and has your scoring been validated against RoMS-registered scorers? Can I see the agreement figures?
• Which milk buyers currently accept your data for their schemes?
• What are the total costs over three years: hardware, installation, subscription and any minimum term?
• Who owns the data, and can I export it to my herd software (Uniform-Agri, Herdwatch, NMR’s systems) without paying extra?
• What happens if the camera goes offline? How quickly do I get a warning, and who fixes it?
• Can I speak to two farms with a parlour like mine, one of whom has used the system for over a year?
• Is there a trial period or a break clause if the alerts don’t match what we find in the crush?

If a supplier won’t give straight answers to the validation and data-export questions, think hard before signing. Those two answers decide whether you’ll still be using the system in year three.