Data science consultancy in Tilburg
We build models using camera footage and measurement data, from the first camera to a model in production. This gave Floating Farm insight into its feeding process.
Using computer vision, Floating Farm gained insight into its feeding process to reduce feed waste
For the floating farm in Rotterdam, we built an end-to-end solution: from installing the cameras to implementing a computer vision model that captures the feeding process.
We see it every day on the shop floor, but how do we measure it?
Much of what happens in a process is observed but not recorded. Cameras and sensors can measure it, and a model turns the images and measurements into figures you can use to adjust the process.
From images, measurements and free text to metrics a team can act on
Making a process visible
For Floating Farm, we installed the cameras ourselves and built the model that tracks the feeding process.
Counting who passes by
For GVB, we used computer vision to establish the actual passenger count.
Classifying free text
For WUA, we built a model that classifies respondents’ open-ended answers by topic.
In production, not in a report
The model runs at set times, with checks that its results remain accurate and an owner who makes adjustments.
Measure what really happens before choosing a model
-
Set up the measurement
Which camera, sensor or source captures what you want to know, and how often.
-
Test the model on real images
Test the model against images and measurements from your own process, alongside what your team sees.
-
Run and monitor
Routine processing, a dashboard or alert for the team, and checks for anomalies.
Can cameras and a model make our process measurable?
A good fit if
Something happens at a fixed location that you currently don’t record, or only record manually, and you want to know how often it happens, how much there is or when it happens.
Not a good fit if
The data is already properly recorded in a system. In that case, a report or dashboard based on that data is the quicker first step.
What determines the scope
How many locations or sources are involved, whether cameras or sensors are already installed, how quickly the results are needed and who will manage it afterwards.
Can you simply install cameras?
Not if people can be identified in the footage. In that case, we’ll agree with your privacy officer in advance what the model sees, what is stored and who has access.
Manual counting, sensors or a camera-based model: which fits your process?
Manual recording
Choose this if you only need to record data a few times a week. A good form and a consistent routine are enough.
Sensors or meters
Choose this if you want to track a single variable, such as weight, temperature or consumption. Our advice: start here if the measurement itself gives you the answer.
A camera-based model
Choose this if you want to see what is happening with things like animal feed, people or products, and measure it repeatedly.
On the factory floor, in the barn or on the road: our models have been running there since 2016
Data Science Lab has been around since 2016 and is a Microsoft and Databricks partner. We’ve worked for Floating Farm, GVB and WUA, and share what we learn at our Techdays.

We start by discussing what you want to measure on the shop floor
Stef Stoelwinder
Business DevelopmentStef is your point of contact. In our first conversation, we discuss which process you want to measure and what is already being recorded.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which process would you like to quantify using images or measurements?
Tell us what happens and what is currently recorded. We’ll reply by email or arrange an initial conversation.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
It builds models that measure, classify or predict something, and makes sure they are used in day-to-day work. This page focuses on images, sensor readings and free text: data that exists but has not yet been turned into numbers.
You can find other applications under AI applications.
No. We work from Westerdoksdijk in Amsterdam. We work online with organisations in Tilburg and the rest of Brabant, and visit sites when cameras or measurements are involved.
None of the case studies here took place in Tilburg; they show what we build.
It uses a model to interpret camera images and recognise what they show, such as an object, a quantity or an action. At Floating Farm, we used real-time image classification to monitor the feeding process.
Read more about the technology under computer vision.
That depends on how much the conditions vary: lighting, angle, season and what appears in the image. We start with a short period of recording and labelling, then check whether the model can distinguish between the different situations.
If a situation proves too rare, we tell you before any further development.
Yes. For WUA, we built a topic classification model that groups respondents’ open-ended answers by subject, making them immediately useful when advising clients.
Read more under text analysis with NLP.
By deciding in advance what the model needs to see and what it does not. Often, the camera can be positioned so that people cannot be identified, or the footage can be deleted immediately after processing, leaving only the counts.
If people are involved, a DPIA is required; see DPIA for AI.
Where the team already looks: a dashboard, an alert or an integration with the system that controls the process. A result that appears only in a report changes little.
Read how we keep models running in production under MLOps.
When a simple measuring device or a well-designed form provides the same answer. Or when no one will use the result to do anything differently.
In that case, we would rather start by helping you record and display what is already there.