Your cameras stay yours. Only the numbers leave the building.
PERCEPTA runs AI detection on the CCTV your city already operates. No new
cameras, no new poles, no procurement of hardware you don't have. A pilot
takes 10 cameras and 90 days to tell you whether it is worth more.
In Europe since 2018ITS World Congress, CopenhagenDOLL Living Lab, DenmarkAutonomous shuttle, Tartu
What crosses your perimeter
Inside your premises
Raw video — never transmitted, never stored by us
Faces & licence plates — blurred into the frame before it is ever encoded
Identities — no plate, no face, no persistent ID is ever written
→
Edge
What we receive
A place — latitude and longitude
A time — to the second
A measurement — "queue, 16 vehicles, 119 m, cleared after 7 min"
This is architecture, not policy. The blur is applied by the video
pipeline before the encoder, so an unblurred frame is never written to
disk in the first place — there is no setting that turns it off and no
file to leak.
Built to the GDPR. No personal data is processed, because none is
ever created: no face, no plate, no persistent identifier, nothing that
can be linked back to a person. We have already completed a full Data
Protection Impact Assessment for a European deployment. The same
architecture satisfies Thailand's PDPA, where we operate today.
How a pilot runs
Week 0
You pick 10 camerasThe junctions you already argue about. We need the existing stream and
somewhere on-site to put a small edge box — roughly one per ten cameras.
Weeks 1–3
We calibrate and connectEach camera is spatially calibrated so a detection lands on the right
metre of the right road, not just "somewhere in frame". Nothing is
installed on your network that sends video out.
Months 1–3
Detections accumulate — free of chargeFlooding, near-misses, queues, wrong-way driving, road damage. Your
officers use the dashboard; residents can see a public version if you
want one. You pay nothing during the pilot.
Month 3–6
You get a number, then you decideWhere the near-misses actually cluster. How long queues really last.
Which complaint was and wasn't borne out by the camera. Only then is there
a conversation about citywide deployment.
What it detects today
FloodingStanding water, with the extent of road covered
Near missesTwo road users who nearly met, with the margin in seconds
QueuesHow many vehicles, how long the tail, how long until it cleared
Wrong-way drivingA vehicle against the flow, and for how long
Road damage & worksPotholes, surface deterioration, lane closures
Citizen reportsResidents report; your officers check it against the camera
What your officers actually do with it
Thousands of detections do not help anyone on their own. What helps is an
answer to a question your team already has. Below are the questions the
screen answers. Your staff open it and read the answer themselves —
none of this is analysis you have to ask us for and wait.
“Is this week worse than last week?”
Every count sits beside the same count for the preceding
period of equal length, with the direction of travel. Near misses,
queues, flooding hours, accidents, wrong-way driving, resident reports.
“Which junction should we fix first?”
Locations ranked by events per observed hour, not by
raw count — so a busy camera doesn't outrank a dangerous one just for
being switched on longer.
“When does it actually happen?”
A day-by-hour heatmap. Scheduling enforcement, works or
crossing-guard hours against it is the cheapest intervention a city has.
“What happened to that complaint?”
Every resident report carries a status — Reported,
Acknowledged, In progress, Resolved — and the resident sees their own
version of it. The queue shows how many are open and how long they have
been.
“Can I put this in front of the council?”
Print to PDF, or export the exact rows on screen as CSV.
The file contains the same rows you are looking at — nothing is
summarised or filtered on the way out.
“Can I just ask it?”
Type a question in plain language and it answers from the
same figures shown on the page. It will not estimate, and it will not
reach for a number that isn't there.
The one no competitor can copy
Two independent records of the same incident
A resident reports a flood. Twenty minutes later a camera detects flooding
fifty metres away. The screen pairs them, and the pair is worth more than
either half.
A report with camera evidence behind it can be acted on straight away
— nobody has to be sent out to confirm it first. And a camera detection
that residents also reported is not merely technically present; it is
affecting people, which is a different priority.
The screen also shows which of the two came first, and by how long. Over a
few months that tells you something a city rarely gets to measure: whether
the reporting channel is adding anything the cameras were going to catch
anyway.
What a city can set in motion
PERCEPTA puts one anonymised dataset in front of three groups who
today never see the same numbers — residents, your own departments, and
industry. That single decision is what turns a deployment into something
larger than a deployment.
Residents
See it free. Published openly, so residents can
see what the city is dealing with and what has been fixed. Those who
report problems add what the cameras cannot see.
Your departments
Act on it. The dashboard above, plus the report
queue, running across every connected camera in the city.
Industry
License it. Anonymised road data from your cameras
is sold to insurers, mapping companies, carmakers and logistics
operators.
Why this matters beyond the dashboard
Only a city can create a supply of this data that
companies will actually build on
What happens on a city's roads is some of the most valuable data there
is, and today it is close to unobtainable. Insurers, mapping companies,
carmakers and logistics operators each spend heavily to approximate it
from the outside — badly, and separately. Price is not what stops them.
What stops them is that no supply exists with the three properties a
business can be built on:
Safe. Anonymised at the camera, before anything is recorded.
No personal data is created, so none can be traded away. A city is
not giving up anything that belongs to its residents.
Fair. Published terms, the same access for everyone,
no exclusivity to anyone.
Durable. The city is the counterparty. The supply does not
disappear because a vendor is acquired or changes its pricing.
We are a private company too — which is exactly why the terms are not
ours to set. The city holds the data rights and decides the conditions
of access, and our agreements bind us to publish those conditions and to
grant exclusivity to nobody. We have deliberately left ourselves no way to
tilt it. A vendor that owns and resells the data itself can say the same
things, but nothing holds it to them, and the city has no say if it changes
its mind.
That is what a company needs before it will commit engineering budget: not
a promise, but a structure where the promise is not the vendor's to break.
The firms that go on to build something operate in your city, and
a wider base of business activity is a wider tax base. The city's
contribution to that is not a product. It is the conditions.
Do we have to replace our cameras?No. If a camera produces an RTSP stream, it can be connected. The whole
premise is that cities have already spent the capital on cameras and are
getting footage out of them rather than information.
Who owns the data?That is settled in the agreement before the pilot starts, and it is
your call. Our model publishes anonymised results in three directions —
to residents, to your departments, and to industry under licence — but a
city that wants a closed pilot gets a closed pilot.
Can we use the speed figures for enforcement?No, and we will put that in writing. Our speeds carry a range-dependent
bias of about 15% that we have not yet resolved, and we mark every reading
as not comparable between cameras. They are good enough to tell you where
to look and not good enough to issue a ticket.
What does it cost?The pilot is free. Beyond it,
pricing is per city plus per connected camera, and we would rather scope
that against your real camera count than quote a number on a web page.
Often the question behind it is where the
funding sits. The anonymised data your cameras produce is licensed to
industry, and that licence revenue can be set
against the operating cost of the deployment, so that continuing the
programme need not rest on a fresh capital allocation each cycle. It is a
funding structure — contracted with the city, on the city's terms, open to
audit — and not a way around procurement. We would rather work it through
against your actual numbers than put a break-even date on a web page.
Who are you, and have you worked in Europe?Avanti R&D — a California company that has built computer vision
for government and industrial customers for ten years.
Europe is where this technology was first shown
publicly. At the ITS World Congress 2018 in
Copenhagen we were a selected demonstrator for the City of Copenhagen,
running a bicycle monitoring system —
cyclist counts and journey times measured between non-overlapping cameras
by anonymous re-identification, at 78.6% re-identification accuracy, with
no imagery ever leaving the sensor.
That led to the DOLL Living Lab in Albertslund,
Denmark, one of Europe's largest smart-city testbeds, under a 2020
Memorandum of Understanding with the Royal
Danish Embassy and the University of Electro-Communications — where we
ran combined traffic and air-quality sensing through 2025. In
Tartu, Estonia, we delivered perception for
an autonomous shuttle programme, with a full GDPR Data Protection Impact
Assessment completed for the deployment.
PERCEPTA is the productisation of that decade of work, and it runs live in
Chiang Mai today.
Citywide deployment in Phuket is covered by a signed MOU.