VisionDesign complete (stages 1–6), not built

FlowCount Lane

Measured wait times for venue lines, from counts alone

Two ceiling nodes count people into and out of a line; the system works out queue length and each person's wait, shows it on a TV and tells the supervisor when to open another register.

Who buys it: Operations or guest-experience manager at a mid-size venue: museum, zoo, aquarium, minor-league ballpark or arena, or university dining.
Why it pays: About $10k a year per concession stand from fewer walk-aways at peak (est.), plus staff hours moved to where the lines are.
Target price: $649 queue pair + $29/node/month (target); single counting node $349 + $15/month
First-minute wow: A TV by the line reads 'Wait about 6 min', and the supervisor gets 'open register 3' two minutes before the line spills onto the concourse. The sensor sends only numbers.

See the CAD design Concept page

FlowCount Lane 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

Airport tenders against Xovis → mid-size venues and campuses that buy without an integrator
$899 stereo sensor, the same as the Axis P8815-2 → single IMX500 node at about $135 BOM, $349
Many stitched sensors per long queue → two count lines and flow conservation, whatever the queue shape
'No images leave' as a claim → AI on the sensor, no encoder loaded, setup view timed and lockable, 24 h packet-capture test
Predicted waits only → measured wait per person from the two count curves, plus Little's-law prediction
Overlap with StoreSense → small-shop tills left to StoreSense

Use case

Seventh-inning rush at Stand 104

Concessions supervisorStand leadFans in line

01Trigger: arrivals at the queue mouth outrun service for three minutes
02Detection: 14 in line, 2.1 served per minute, predicted wait about 7 min
03Alert: the supervisor gets 'Stand 104: open register 3'; the stand TV shows the wait
04Action: a third register opens, service rate rises and the alert clears

Outcome: The line shortens before fans give up; about $150 a game recovered per stand (est.)

How it works

Architecture

01
Sense: Sony IMX500 runs a top-down person detector on the sensor; the Pi Zero 2 W tracks and counts line crossings
02
Decide: the wait engine computes queue length, service rate, predicted and measured wait, with drift resets
03
Send: count deltas every 5 s over MQTT/TLS; no images, no identities
04
Tell: TV wait page, public JSON feed, supervisor alerts, Monday report

Hardware

Real parts, buyable today

Bill of materials: about $135 per node at retail qty (est.).

Block	Part	Why this part
Sensor + AI	Raspberry Pi AI Camera (Sony IMX500, 66° × 52°)	Detection on the sensor chip; $70 launch price
Host	Raspberry Pi Zero 2 W	$15; tracker and line logic at about 1 W
Power + network	802.3af PoE splitter (5 V) + USB Ethernet adapter	One cable at the ceiling, about 3.3 W at the switch
Housing	140 × 120 mm printed PETG base and cover, T-bar clip, 15° wedge	Drop-ceiling or hard-ceiling mount; side view for wide entrances

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

FlowCount Lane 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

Top-down detector accuracy on the IMX500, and its Python conversion toolchain the lab doesn't have
Lines three abreast and strollers break line counts without stanchions
Venue and university purchasing may still be slow
The Pi Zero 2 W has no secure boot; a stolen node's key must be revoked

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