Service
Computer vision
We build systems that interpret images and live video, from measurements in a phone photo to tracking across several cameras. Models are tested against your footage and site conditions, then connected to the tools your team uses.
- Real time
- live video at the frame rate your site needs
- mm accurate
- measurement from calibrated photos
- Any camera
- phones, CCTV, RTSP streams and depth cameras
Where we can help
- Staff do visual checks by hand and results vary between people.
- You have hours of CCTV footage and no way to search or measure it.
- An off the shelf model works in the demo but not in your site.
- You need to follow people or objects across several cameras.
What you get
Detection and segmentation
Custom models trained on your images, with honest accuracy reports.
Multi-camera tracking
Stable identities across views, occlusion and re-entry.
Measurement and pose
Sizes in millimetres, body joints, angles and movement.
Live alerts and dashboards
Events pushed to email, SMS, PLCs or your own app.
How we run it
From assessment to delivery.
Week 1
Site and camera review
We look at your cameras, light, angles and hardware, and collect sample images or video.
Week 1 to 3
Data and labelling
We label a first dataset, often with model assisted tools to save time, and set aside a fair test set.
Week 2 to 6
Train and measure
Custom models trained and tested on your conditions. You get accuracy per class, not one flattering number.
Week 5 onwards
Deploy and tune
Running on your servers, cloud or edge devices, with alerts, dashboards and a plan for retraining.
Where to start
Vision feasibility test
2 to 3 weeks, fixed price
Send us footage or photos from your site. We train a first model and show you what accuracy is possible there.
- Review of cameras and setup
- First labelled dataset
- Baseline model with accuracy report
- Hardware and cost advice
When we would advise against it.
If one of these applies, we will tell you in the first call.
- A barcode, sensor or scale already gives you the answer reliably.
- You cannot share sample images or footage from the real site.
- The check happens a few times a week. Doing it by hand costs less.
Typical projects
Common applications.
- Smart checkout and loss prevention in retail
- Size grading in aquaculture and food
- Vehicle and safety monitoring on production lines
- Sports and physiotherapy motion analysis
- Document detection and OCR preparation
- People, vehicle and licence plate counting
Tools we use
Tools that fit your project.
We choose tools around your existing systems, data requirements and operating costs.
YOLO · RT-DETR · SAM 2 · ViTPose · ByteTrack · OpenCV · PyTorch
Related work
Related projects.
Kiss and Makeup
Following products from shelf to checkout
A multi-camera proof of concept for Kiss and Makeup, connecting shelf pickups, checkout and alerts at the exit.
- Shelf to checkout
- items followed across cameras
- Alert at the exit
- with a photo, by email and SMS
- Phase one delivered
- as a proof of concept in the shop
Retail · Upper Hutt, New Zealand
US car wash equipment maker
Detecting vehicle rollback in car washes
A camera system for a US car wash equipment maker that detects vehicle rollback, signals the conveyor to stop and records the event for review.
- Conveyor stops on its own
- when a vehicle rolls back
- Every event recorded
- with video, for staff to review
- In production since 2024
- with regular releases
Automotive and equipment · United States
Mussel App by FutureLab
Measuring mussels from a phone photo
An image-processing API for Mussel App by FutureLab. A calibrated photo becomes measurements the team can review in millimetres.
- Millimetres from one photo
- for every mussel on the plate
- No special camera
- a phone photo with a printed scale
- Runs as an API
- used by the team’s iOS app
Aquaculture and agriculture · New Zealand
Questions about this service.
How many images do we need?
It depends on the task. Many projects start well with a few hundred labelled images per class, and we tell you early if more are needed.
Can you use our existing cameras?
Usually yes. We check resolution, angle and light first, and only suggest new hardware when it clearly pays off.
What about privacy when filming people?
We design for it from the start: processing on site, blurring faces, and storing events instead of video where possible.
What would you like to build?
Tell us where your project stands. If AI is not worth it for you, we will say so.