Skip to content
IEMI Labs
All product designs
VisionDesign complete (stages 1–6), not built

ParkSense AI

Find the cars that didn't pay on a gateless lot, and audit what the lot really earns

Counts every car on a small pay lot and matches it to the payments taken, giving the attendant a live 'unpaid now' list and the owner a monthly leak audit.

Who buys it
Independent operators of 1–10 surface lots, and landlords who lease a lot to an operator for a share of revenue.
Why it pays
A 60-space event lot that is 10% unpaid loses about $9,000 a year (est.). A landlord on a 60% share loses about $3,600 a year for every 5 points of under-reported sessions (est.). The real rate is measured in the pilot before it is quoted.
Target price
$1,190 site kit + $49/site/month (target); audit-only tier $29/site/month
First-minute wow
The attendant's phone shows '3 unpaid, longest 41 min: grey SUV, row C', each with an entry photo, and clears itself when a late payment lands.
See the CAD designConcept page
ParkSense AI concept: the idea and where each part goes
The concept. The CAD design is below, under Printable CAD.

What the research found

And how this design answers it

  • Enterprise CV against funded incumbents → sold to independent lots and landlords who buy with a card, not through RFPs
  • Plate-read software is already about $12–55 per camera per month → the read is included; the product is matching cars against payments plus the landlord audit
  • LPR cameras cost $500–1,300 → one ~$250 varifocal camera per entry lane, with OCR on the shared Edge Hub
  • Existing CCTV rarely reads plates → existing cameras only count occupancy; plate reading needs our specified lane camera and aim
  • False 'unpaid' flags hurt paying customers → confidence threshold, fuzzy matching, a stale-payment-data guard, and a person confirms before any action
  • Plate privacy → read at the lane only, hashed, clear text leaves the hub only for listed cars, 30-day purge and a posted notice

Use case

Concert night on a 60-space lot

Lot attendantOperatorDrivers

  1. 01Trigger: a car enters at 18:20; the lane camera reads its plate
  2. 02Detection: after 15 min grace the hub finds no matching pay-by-phone or kiosk payment
  3. 03Alert: the car appears on the attendant's 'unpaid now' page with an entry photo and row hint
  4. 04Action: the attendant finds it and leaves a pay-now notice under the lot's posted policy; a late payment clears it automatically

Outcome: The attendant works 5–10 real cases instead of checking 60 dashboards; about $4,500 a year recovered on 100 event nights (est.)

How it works

Architecture

  1. 01

    Sense: one LPR lane camera and 1–3 overview cameras on the lot, PoE to the Edge Hub

  2. 02

    Decide: plate detection and OCR on the Hailo-8; a Node.js matcher checks plates against payment data every 60 s

  3. 03

    Send: MQTT over TLS, events only; clear plates leave the hub only for listed cars

  4. 04

    Tell: attendant phone page, operator night view, monthly landlord audit PDF

Hardware

Real parts, buyable today

Bill of materials: about $850 per site with 1 lane and 1 overview camera, including the shared hub (est.).

BlockPartWhy this part
HubIEMI Edge Hub (Raspberry Pi 5 8GB + AI HAT+ 26 TOPS Hailo-8)Shared with LotWatch; runs plate OCR locally, no video upload
Lane camera2 MP varifocal bullet with manual shutter (IPC-HFW5241E-Z12E class)Narrow angle and fast shutter put 130+ px on each plate at up to 15 mph
Overview cameraReolink RLC-510A 5 MP PoEOccupancy and vehicle colour/type for $40–60
MountPrinted ASA pole saddle + adapter plate, stainless band clampsFits 50–115 mm poles without drilling the operator's pole

Mechanical

Printable CAD

4 parts, generated from code and checked as closed solids. Pick a part to inspect it, or download the STL files to print.

See the CAD parts render
ParkSense AI printable parts, rendered from the CAD
Render of the printable parts from the actual CAD. Not yet printed or photographed.

Loading the CAD…

Real CAD · drag to rotate · scroll to zoomReal CAD · swipe sideways

Open risks we are working on

  • US plate OCR accuracy on our own pipeline is unproven, especially at night (fallback: licensed Plate Recognizer at a higher fee)
  • Payment data access depends on each pay-by-phone vendor; the CSV path makes the unpaid list slower
  • A false 'unpaid' flag acted on by staff; mitigated, never zero
  • Large operators and vendors moving down-market to small lots

Design stage only: nothing on this page is for sale. Specifications and prices are targets.