Edge AI for real-world perception
Turn your cameras into real-world intelligence
Peregrine reads the scene in front of the camera in real time, on the device. For road safety, driver assistance and site security.
- Under 85 ms, on the device
- No new hardware required
- Anonymised before storage
Working with
From detection to action
See it. Understand it. Act on it.
The whole sequence takes 85 milliseconds and never leaves the camera. Watch the feed stop.
Inside the camera · no connectivity required
FRAME 001 · HELD cyclist · entering lane
closing 4.1 m/s · gap 1.9 s
driver gaze off-road
The situation, not the object: trajectory, gap and risk.
- Vehicles
- Vulnerable road users
- Road assets
- Environment
All of that happened on the device. Only the result ever leaves it.
Faces and plates are blurred before any of it is viewed.
at 85ms, frame after frame.
Why now
Cameras are everywhere. Understanding isn't.
From cars, trucks and forklifts to drones, bodycams and public spaces, cameras already capture the physical world at enormous scale. Most of them can detect. Very few can interpret a situation.
The architecture
Small enough to live inside the camera
Objects, depth, motion and lane geometry share one backbone and one pass. Four separate models would need more power and more time than a camera has.
The architecture
One shared backbone, several task heads. Four jobs in a single pass keeps inference under 85 ms on a camera's own processor. Four models in sequence would fit in neither the power budget nor the time.
Three outputs, one pass. Context, depth and lane geometry come out of the same forward pass.
Three conditions, one model. A Berlin junction, an alpine blizzard and a motorway, with no retuning between them.
Where we fit
Make your products and services context-aware
Peregrine runs on the device and hands your backend structured, anonymised events: what happened, where, and what made it risky.
Event delivered to your platform
typetailgating + distraction
risk92 / 100
gap0.9 s
mediaclip, faces & plates blurred
- You own the hardware.
- We add our perception software.
- Together, we transform your platform's UX in the age of AI.
Built for connected camera platforms
Designed for deployment
Runs on silicon you ship
We quantise the model to the compute already in your architecture.
- Quantised to your NPUCompiled for the chip you chose, within the power and thermal budget you have.
- Retrofit over the airDeployed to vehicles on the road, whatever their make, model or age.
- SDK or APIEmbed it in firmware, or take structured events over REST and webhooks.
- Works without connectivityEverything decisive happens on the device. Coverage gaps cost you nothing.
Transportation
Real-world industry solutions
Understand the road, the driver and the context.
Hazards it detects
- Red-light violation
- Speeding
- Adverse weather
- Slippery road
- Size & weight restriction
Webfleet · a Bridgestone company · Sept 2025
Visual Intelligence, inside Webfleet Video
A paid upgrade, retrofitted to commercial vehicles in service. Webfleet serves more than 50,000 businesses worldwide.
“…detecting and contextualizing road events and risks as they happen.”
Jan-Maarten de Vries, President, Fleet Management Solutions, Bridgestone
Alerts in the Linqo dashboard
- Speeding
- Tailgating
- Fully anonymised
Linqo GmbH · Berlin & Vilnius · May 2025
Peregrine Vision on Linqo’s dashcams
A separate company and platform running the same engine, on dual-camera hardware Linqo customers already have.
“Our customers want systems that just work — and work together.”
Max Donders, Managing Director, LinqoPeregrine Data
Turn every kilometre driven into insight
Once a camera understands what it sees, every route becomes a survey. The driving was happening anyway.
For your platform
Improve safety features, context services and maps with data your own devices produced.
For your fleet customers
Safety and risk intelligence from perception deployed in their vehicles.
Beyond the fleet
Road and infrastructure insight for cities and road authorities. Anonymised, geo-tagged and dated.
Built-in privacy
Privacy starts at the edge
Faces and licence plates are blurred automatically, at petabyte scale. 99% of the visual data is filtered out on the device, and what remains is encrypted.
How the anonymisation works
Evidence
AI research and deep tech programs
Selected by leading deep tech programs globally, and collaborating with leading Universities in Germany in Robotics and Edge AI.
Public sector & research deployments
Hamburg
Potholes found before they spread
Fleet vehicles on their normal routes surface road wear and infrastructure faults, so the city can act while repairs are cheap.
With the City of Hamburg
Read the case
Landshut · 5-SAFE
School zones that warn drivers
5G sensors and adaptive signage watch pedestrian and vehicle movement, and fire only when the risk is real.
With T-Systems, Landshut University and the City of Landshut, backed by the BMDV
Read the case
Gaia-X 4 AMS
The scene, before arrival
Drone footage with detection overlays streamed into fire brigade vehicles, so crews saw the site before they arrived.
With the Gaia-X consortium and local fire services
Read the caseDeployment footprint: Berlin · Hamburg · Landshut · European fleets.
Research partners
In early 2023, we hired the first Ph.D. student conducting cutting-edge research at the intersection of AI and sustainability as part of a joint program between HPI and Peregrine.
Professor Ralf HerbrichHasso Plattner InstituteWe have successfully conducted joint research on Self-Supervised Contrastive Learning in Traffic Scenes, experimenting with new approaches for Instance Segmentation using unique data provided by Peregrine.
Professor Cyrill StachnissUniversität BonnWe worked with Peregrine and Seagate to experiment with real-time data logging and Peregrine’s smart data campaigning system for real-time data annotation using our self-driving vehicle “MadeInGermany”.
Professor Daniel GöhringFreie Universität BerlinPrograms & consortia
Peregrine Labs · the technology engine
Advancing the state of the art in edge perception
A model that only sees easy scenes stays good at easy scenes. Every camera in the field flags what it found hard.
- 01
The fleet finds the edge cases
An unfamiliar sign. A barrier at dusk. Flagged on the device.
- 02
Campaigns ask for what is missing
Ask the fleet for wet-night junctions or a sign variant. The frames come back annotated and anonymised.
- 03
Updates go back over the air
Improved models reach devices already in service. Yesterday's edge case is today's ordinary scene.
Beyond the road
Wherever a camera has to think for itself
Roads are where Peregrine Vision is proven. A drone, a bodycam and a forklift pose the same problem: limited power, no server, a decision needed now.
Newsroom
What we're publishing
Webfleet and Peregrine.ai collaborate on Visual Intelligence
Bridgestone's fleet platform adds contextual perception as a paid upgrade, retrofitted over the air.
Read the release Press releasePeregrine.ai and Linqo partner on integrated video telematics
Edge computer vision in an established fleet platform, for commercial fleets across Europe.
Read the release Fleet managementWhy your video telematics strategy is trapped by G-force
Accelerometer triggers catch the harsh brake. They never catch the near miss that needed no braking.
Read the post LabsVisual SLAM vs. LiDAR: spatial intelligence without the hardware tax
Spinning laser arrays are the reason spatial data has stayed expensive. They are not the only route to it.
Read the postPeregrine in the press
- Handelsblatt
- WirtschaftsWoche
- Berliner Morgenpost
- The Pioneer
Your cameras see the world.
Now help your platform understand it.
Evaluate Peregrine Vision on our or your own hardware.