VisionDesign complete (stages 1–6), not built

ShelfScan Gap

Empty-shelf tasks with a photo, for the bays that make the money

Battery cameras under the shelf edges photograph the shelf across the aisle every 15–30 minutes; the store's IEMI hub finds empty rows and sends staff a restock task with a photo, then confirms the restock.

Who buys it: Owner or operations director of an independent grocer or small chain (1–10 stores).
Why it pays: About $7.5k a year per store (est.) from gaps fixed hours sooner on 30 money bays plus less shelf-walking, against $1,188 a year of subscription.
Target price: $1,490 store starter (hub + 12 cameras) + $99/store/month (target)
First-minute wow: Twelve cameras on magnets in a morning, and the first 'Bay 4-C shelf 3, about 40% empty' task with a photo before lunch. No product catalogue needed.

See the CAD design Concept page

ShelfScan Gap 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

Funded chain incumbents (Focal, Captana, Trax) → sell to independents, 20–40 money bays, published price, no RFP
SKU catalogue cost → v1 detects gaps only; the task photo shows staff what is missing
$299 4K AI cube needing shelf power → ~$32 battery still-camera, AI runs once on the store hub
Shoppers in frame → any shot with a person over the shelf is deleted on the hub and retaken
Price-tag OCR and freshness claims a camera can't prove → dropped
Restock never checked → the next photo verifies it and logs time-to-restock

Use case

Saturday gap in the beverage aisle

Grocery managerStockerShelfCamEdge Hub

01Trigger: the 14:00 photo of bay 4-C shows the 12-pack row 45% empty, confirmed at 14:15
02Detection: the hub scores each 5 cm cell of the row and opens a gap after two low shots
03Alert: the stocker's phone shows the row photo and how long it has been empty
04Action: restock, tap Done; the next photo verifies the row and logs 22 minutes to restock

Outcome: The gap is fixed during the peak, not at the 17:00 shelf walk; roughly $15k a year of sales protected in one category (est.)

How it works

Architecture

01
Sense: ShelfCam (XIAO ESP32S3 Sense, OV3660) wakes every 15–30 min and posts one JPEG to the hub
02
Decide: the hub rejects shots with people, then scores shelf-row fill with hysteresis
03
Send: row states and gap events over MQTT/TLS; only people-free row crops for open tasks leave the store
04
Tell: store app tasks, a hot-bay re-check after restock, and a Monday worst-bays report

Hardware

Real parts, buyable today

Bill of materials: about $32 per ShelfCam and about $190 for the Hub Lite at retail qty (est.).

Block	Part	Why this part
Camera + MCU	Seeed XIAO ESP32S3 Sense (ESP32-S3R8, OV3660 3 MP)	$13.90, 8 MB PSRAM holds a full 2048×1536 JPEG, LiPo charger on board
Battery	Adafruit 2000 mAh LiPo with protection	About 300 days at a 30-minute interval (est., bench test T1–T3)
Mount	Printed hinge + 4 × N52 12 × 3 mm magnets, steel tether	No tools on steel gondolas; tilt ±30°
Store hub	IEMI Edge Hub Lite (Raspberry Pi 5 4GB, no AI HAT)	About 0.02 images/s needs no NPU; full hub if the store runs other app packs

Mechanical

Printable CAD

5 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

ShelfScan Gap 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

XIAO Sense sleep current above 100 µA would halve battery life and force a custom board
The gap heuristic may confuse pale packaging with an empty back panel until the classifier is trained
Independents may not pay $99 a month; the 30-day pilot tests it
Firmware toolchain (ESP-IDF) is not installed in the lab yet

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