Embedded & IoT
Computer vision on the camera, not in a data centre
Detect stock gaps, queues, safety breaches and anomalies in real time, on edge devices you own, with privacy designed in.
- Cameras you have, or need
- The decision the camera should make
- Privacy rules
- Sample footage
- On-shelf availability and planogram checks
- Queue, footfall and dwell analytics
- PPE and safety-zone monitoring
- Defect and anomaly detection on lines
- Model training on your own footage
- Edge deployment and fleet updates
- Interfaces
- Python
- PyTorch
- ONNX
- OpenCV
- C++
- IP cameras (RTSP / ONVIF)
- Applies to
- Stores · Warehouses · Production lines
- Engagement
- Vision pilot · Rollout pod
In plain words
Cameras that understand what they see, like an empty shelf, a long queue or a missing safety helmet, and alert the right person straight away.
- Who it's for
- Stores, warehouses and factories that already have cameras but only use them to record.
- How to start
- A pilot on a few of your existing cameras.
Words on this page, explained
- Computer vision:
- Computer vision: software that understands what a camera sees, such as an empty shelf or a missing helmet.
- Edge AI:
- AI that runs on the device or camera itself, so video and data don't have to leave your site.
Most vision projects stall between a good demo and a site that works every day. We build for the site, with its lighting, camera angles and network limits. Models run at the edge, so decisions take milliseconds and video never has to leave the building. Alerts go where your teams already work: the WMS, the store app, Slack or a siren. Face recognition is off by default and only enabled after a legal review.
Computer vision · produce grading
Fair grades, decided by a camera.
Grading by eye is slow and leads to arguments. A camera over the belt checks size, colour and marks on every piece the same way, and sorts it to the right crate.
Packhouse belt · camera 1
Last piece checked
Over-ripe: send for pulp
Size 82 mm · no marks · over-ripe
What you’re seeing: fruit moving under a camera. Each piece gets a grade and a reason in plain words, then goes to its crate. Switch the buyer and every crate is re-counted to that buyer’s rules.
For engineers
Top-down camera with controlled lighting; an edge AI model segments each piece, measures diameter, colour distribution and surface defects, and maps them to the buyer's grade card. Results drive the diverter and are logged per lot with images for audit. Models are trained on your own produce.
Fruit 16: Grade C. Over-ripe: send for pulp.
Why teams call us
- 01
Cameras that record but don't understand
Footage nobody watches doesn't prevent anything.
- 02
Privacy concerns blocking projects
We process at the edge and keep only events and counts, not faces.
How we work
From first workshop to running in production.
- STEP 01
Define the event
What should the system notice, and what happens next?
- STEP 02
Collect and label
Representative data from your sites, with privacy controls.
- STEP 03
Train and optimise
Models sized for edge hardware and your accuracy target.
- STEP 04
Deploy and monitor
Edge devices, alerts, dashboards and drift monitoring.
What you get
Deliverables, not slideware.
- Edge inference devices
- Configured and monitored
- Models and evaluation
- Accuracy reports on your data
- Event pipeline
- Alerts into your workflows and dashboards
Technology
No stack limits. Right tool for the problem.
Models
Edge
Pipelines
Need something not listed? We hire for it or bring in a specialist. The problem picks the stack.
Shelf and safety events in seconds
Not after the shift report
Privacy by design
Counts and events, not identities
Computer vision & edge AI: common questions
Do you do facial recognition?
No. Our retail and industrial solutions are designed around anonymous events and counts.
Can you reuse our existing cameras?
Often, yes. We assess the angle, resolution and network first.
Engagement models
Ways to work with us
Vision pilot
one site, one use case, measured against a baseline in 4–6 weeks
Rollout pod
models, devices and integrations for every site
Where this sits in the stack
Six blocks on one die. One team across all of them.
We work across .NET, C, C++, MFC, Go, Java, Python, Node.js, React and Spark, on GCP, AWS and Azure, plus embedded C on RTOS and embedded Linux. Pick a block to see where it fits.
Hover, tap or tab through the blocks. Lit copper shows which blocks work together.
Block · AI
Models at the edge and in the cloud.
- Python
- PyTorch
- OpenCV
- ONNX
- Computer vision on edge devices
- Model training on your own footage
- Forecasting and anomaly detection
- LLM copilots over your documents
Where it shows up
- Computer vision (this page)
- Innovation Labs
- Retail technology
Scope your Computer vision & edge AI project
Tell us the problem, the constraints and the deadline. A senior engineer replies within one business day.