CONTROL ROOM / LIVE CHECKOUT INTELLIGENCE

Every frame becomes a decision.

Overhead frames → person & basket detection → basket-aware wait prediction → lane signal. The model keeps learning from real checkouts.

Shoppers in queue

—Detected across all lanes

Estimated items

—From basket fullness classes

Fastest estimated wait

—No open lanes

Checkouts completed

—0 with timing feedback

Live lane overview

Fast Moderate Busy

Lane 01

—waiting for server

Lane 02

—waiting for server

Lane 03

—waiting for server

Lane 04

—waiting for server

Overhead vision · —

No frame analyzed for this lane yet

OFFLINE

NO FRAME

Bring your own queue photo

Drop or paste a photo of a checkout line with shopping carts or baskets — overhead, CCTV, or phone angle. The vision engine counts the shoppers, rates every basket, and turns it into a wait estimate for .

JPG, PNG or WEBP · Ctrl/Cmd+V pastes from the clipboard · nothing leaves your local network

YOLOV8 PERSON DETECTION · BASKET FULLNESS PER SHOPPER

ANALYZE INTO

CAPTURE SOURCE

DETECTIONS → SERVICE TIME

0shoppers in —

No shoppers detected.

Lane wait estimate—

Per-shopper estimate = … seconds, where items come from the basket fullness class. No connection to the vision server.

Online learning · prediction accuracy

predicted_sec = … · updated by SGD after every real checkout (0 synthetic warm-up + 0 live samples)

first 10 —%last 10 —%

Transparent to judges: the warm-up curve replays synthetic checkouts so the mechanism is visible from the first minute. Every “Done” or “Teach the model” action is a real SGD step on the same regressor — no code changes needed when real store data arrives.

— · Detected queue

0 shoppers

SHOPPER	FULLNESS	CONF.	ITEMS	SOURCE	REMAINING

No shoppers in this lane.

Signal thresholds: green ≤ 2 min · amber ≤ 4 min · red > 4 min, computed by the vision server from the sum of per-shopper estimates. “Done” measures the real service time of the front shopper and feeds it back into the regressor. API: http://127.0.0.1:8000