Solutions / Catalogue / Edge Video Analytics

Edge Video AnalyticsScenario

Smart Retail Store

Traffic camera and shelf AI box

An in-store vision system with three jobs: footfall counts and heatmaps, product recognition at checkout and on shelves, and face-based door access for unstaffed stores. Cameras feed a local reComputer or an all-in-one reCamera, and no video is sent to the cloud. It suits retailers and integrators who want store analytics without paying for cloud video processing.

Retail

Talk to an engineer See the hardware

Inside an Indian grocery store, shelves packed with packaged goods above sacks of grain — Photo: Balurbala, CC BY-SA 4.0, Wikimedia Commons. Shows the setting, not this installation.

84.67%: Top-1 product recognition on a public 81-class grocery dataset with 8 photos per item (dataset, not a live store)
0.66 s: Face unlock signal time on reCamera Pro with 1,000 people enrolled
22.3 ms: Median checkout detection time on J4012, with no dropped frames in 2,956
924 ms: Median shelf recognition time on RK3588, detection to result

Results reported by the design's authors on their own test setups. Yours depend on the site, so we confirm them in a pilot.

The problem

Why it matters.

Beam counters miss dwell and turnarounds, cloud video is costly and intrusive, barcode checkout and shelf walks are slow, and keycards or QR entry fail for unstaffed doors.

The flow

From signal to action.

01SenseStore camerasExisting RTSP cameras or reCamera units watch the floor, the checkout belt, the shelves and the door.
02DecideEdge hostA reComputer runs people tracking, product recognition or face matching on each stream.
03RecordLocal dashboardCounts, dwell times and heatmaps go to a dashboard on the store LAN, with no identities kept.
04ConnectPOS linkRecognised basket items and shelf status are passed to the POS and console over local APIs and MQTT.
05ActDoor lockA matched face triggers a relay that releases the electronic lock, and the entry is logged.

Deployment architecture

How it is wired.

The devices on site, what carries their data, where the decision is made and where the result goes.

01SenseOn site

reCamera 20021 per counted area
reCamera Pro1 per door
IP camera (RTSP)1 per task

02ConnectCarries the data

RTSP
GPIO and relay

03DecideRuns the model

reComputer R1125-101 per store
reComputer RK35881 per task
reComputer J40121 per checkout
reComputer J30111 per store

04Your systemsWhere results land

MQTT
Docker
POS (local API)
Seeed solution CLI

05ActWhat happens next

Door lock

How it works

Step by step.

01Mount each camera for its task: high and looking down for footfall, directly over the belt for checkout, facing the shelf for audits, and at face height at the door.
02For heatmaps, mark four matching points so each camera view lines up with the floor plan.
03Register each product with 3 to 8 photos. No model retraining is needed, and the library can be rolled back.
04The host lists basket items at checkout and marks each shelf slot as fine, empty, wrong item or unknown.
05At the door, the camera checks the face is live, matches it to the local list, unlocks and writes a log entry.

The hardware

What goes on site.

Items we stock can go straight in the bag. Everything else we source and quote with the installation.

reCamera 2002All-in-one footfall camera for small areas1 per counted areaSourced on quote
reComputer R1125-10Dashboard backend for reCamera footfall1 per storeSourced on quote
reComputer RK3588Footfall tracking host or shelf recognition host1 per taskSourced on quote
reComputer J4012Checkout basket recognition1 per checkoutSourced on quote
reCamera Pro4K door camera that does face recognition on board and drives the lock relay1 per door₹32,989
reComputer J3011Face recognition host that can serve several doors1 per storeSourced on quote
IP camera (RTSP)Customer-supplied video source for each task1 per taskSourced on quote

Before you install

Footfall cameras mounted high and angled down, covering about 3 to 5 m.
A checkout camera directly above the belt with items fully visible, and shelf cameras facing the shelves.
Door cameras at face height where people stop, away from backlight and glare.
A store LAN plus a separate computer that can run Docker for the checkout and shelf consoles.
An installer to supply and wire the relay, electromagnetic lock and lock power.

Good to know

Footfall speeds were measured on single streams, so test before sizing a host for many cameras.
Product accuracy was measured on a public grocery dataset and drops with fewer enrolment photos. Real shelves with similar packs may score lower.
Face access struggles with backlit doorways and glare, and staff face data is personal data that needs consent and a retention policy.
It is vision only. Counter conversations need a separate audio solution.

How we deliver it

Paper to site, one team.

01Site surveyWe walk the site, or review photos and drawings, and agree what to measure and where each device goes.
02PilotOne lane, one line or one room first. We tune it on your conditions and agree what good looks like.
03Roll outWe supply the hardware in India with GST invoices, install it, and connect it to the systems you already run.
04SupportSpares, remote checks and model updates, with one team to call when something changes.

Questions

Asked before buying.

Is customer video uploaded anywhere?

No. Video is processed in the store, and footfall keeps only counts and times, not identities.

How do we add a new product?

Take 3 to 8 photos of it and register it in the console. No retraining is needed.

Will the door still open if the internet is down?

Yes. Matching runs locally and the last synced staff list is kept offline.

Want this on your site?

Tell us the site and the problem. We come back with a pilot plan and a priced hardware list.

Book a call with an engineer Write to us

Want this on your site?

Industrial Visual Inspection

Parking Lot Management

Elderly Care Monitoring

Want this on your site?

Industrial Visual Inspection

Parking Lot Management

Elderly Care Monitoring