A parking operator was running multi-level structures blind — no live view of which bays were free, which reservations showed up, or which cars sat where they shouldn’t. Axlume turned the existing CCTV into a computer-vision system that sees every bay in real time, then built the driver app and operator consoles on top of it.
Our client operates multi-level parking structures and needed them to stop being operational black boxes. Staff couldn’t see live occupancy, drivers circled floors hunting for a space, and enforcement meant a person walking the levels hours after the fact. Axlume was brought in to make the structure legible — to operators in real time, and to drivers before they ever drive in.
The work spanned a computer-vision pipeline running on the building’s existing cameras, a two-tier operator console, and a customer-facing driver app — one system where what the cameras see drives everything the apps do.
Occupancy lived in people’s heads. There was no live, per-bay view of the lot — only a rough count, updated by whoever last walked through. Managers couldn’t tell a free bay from a wrongly-parked car from an unbooked one.
Drivers felt the same blindness from the other side: circling levels for a space, never sure a “reserved” bay would be free on arrival. And enforcement was manual — slow, inconsistent, and impossible to scale across multiple floors and sites.
Every question that mattered — where’s the space, did the booking show up, is that car allowed to be there — required a human to go and look.
The obvious route to “smart parking” is hardware in every bay. We did the opposite — we built a vision layer on the CCTV the structure already runs, so the intelligence is software, not a construction project.
We mapped the real operational gaps — what staff couldn’t see, what drivers couldn’t trust, and where manual enforcement broke down — against the camera coverage, the floor layout, and the booking flow.
We modelled the lot the way the cameras would: floors, zones, blocks and individual bays, accessible spaces included. That model became the single source of truth every surface reads from.
Object detection locates and classifies each vehicle per frame; OCR reads the plate to tie a car to its booking; an occupancy layer reconciles what the cameras see against what the system expects — so free, wrong, and unbooked are all distinguishable, automatically.
A FastAPI service layer sits between the vision pipeline and the apps; MongoDB holds the lot model, live sessions, and the event stream off the cameras. Detection is decoupled from booking, so the driver app stays fast while the heavier CV work runs against the feeds.
Delivered as a working platform — iOS driver app, web operator console, and a super-admin layer — region-agnostic from day one, with configurable currency and localized payment rails.
The vision layer in action — detecting vehicles and resolving per-bay occupancy in real time, tracking riders as they enter and park. This is the proof the rest of the platform is built on.
Detection, plate recognition and occupancy reconciliation running on the structure’s existing cameras — turning passive feeds into a live, queryable model of every bay.
Every bay rendered free, occupied or reserved, accessible spaces flagged, each zone tied back to its camera — so a manager reads the whole structure at a glance instead of walking it.
Because the system knows where every car should be, it flags the ones that aren’t. A vehicle outside its bay or in a restricted area triggers a timed notice to move before the session is marked non-compliant — automatic, fair, and logged.
Booking, live sessions, extend-time, payment and attendance — with the reservation ledger reconciled against real occupancy, so “attended” means a real car was actually in the bay.
A super-admin view for operators running many lots — capacity, zones, pricing and users across the estate — and a lot-manager view for staff running a single site’s floor.
The simple side — find a nearby space, pick an exact bay, confirm plate and time, pay, and extend from the phone, with the same compliance engine warning drivers in time to avoid a penalty.
Drivers only ever touch the easy side — find a space, pick an exact bay, pay on local rails. The same compliance engine that powers the operator console is the one that warns them in time to avoid a penalty.
The lot went from a black box to a live system. Operators got a real-time, per-bay view they never had; drivers got wayfinding and self-service they can trust; and enforcement became automatic instead of manual.
What changed operationally:
Detection accuracy and deployment metrics are held under NDA. Contact us directly for a detailed brief.
Real computer vision in production is the hard part — not a model in a notebook, but detection, plate recognition and occupancy reconciliation running reliably on live camera feeds, under real lighting, real angles, and real traffic.
We kept it hardware-light on purpose: the system runs on the cameras the structure already has, so “smart parking” is a software deployment, not a construction project. The detection workload is decoupled from the booking flow, so the consumer app stays responsive while the CV work runs against the feeds.
And it was built region-agnostic from the ground up — a configurable currency layer and a payment stack supporting both European card rails and South-Asian mobile wallets, so the same platform deploys across markets without a fork. The result behaves correctly under real operating conditions — not just in a screenshot.
You’ve seen our standard of engineering. Now, let’s build yours. From high-impact websites to complex enterprise applications, we specialize in transforming ambitious ideas into elite-grade digital reality. Whatever the platform, we bring the precision your vision demands.