Vision-based card recognition

Every card call
accurate, instant, provable

EWIN AI VISION reads the felt from an overhead IP camera, computes points and outcome automatically, and pushes the result to the dealer tablet, the roadmap boards and your back office. Your floor needs one ingest server — and no change to the table.

In production on live baccarat tables; other card games are in development.

99.9% Recognition accuracy
<300ms Result latency per round
24/7 Continuous operation
29+ Card games supported

Every second and every dispute at the table is a cost

Table yield comes down to pace and trust. Every manual step in the round is a source of error and expense.

Manual calling and entry errors

A dealer's call, a surveillance re-entry — one mistyped card means an incorrect payout, a complaint and a refund, then hours of footage review to prove what happened.

Pace caps your rounds per hour

Rounds per hour set table yield directly. Waiting for manual confirmation, entry and reconciliation adds seconds to every single round.

Disputes take too long to close

When a VIP questions a round, someone has to pull and scrub timeline footage. It is slow, and it is felt directly in the guest experience.

No structured round data

Without a clean record of every card in every round, there is no roadmap analytics, no table efficiency review, no anomaly detection and no risk control.

Not a camera — an operable recognition service

Capture, streaming, recognition, settlement and integration are delivered and owned end to end. Your floor supplies IP cameras, one ingest server and a tablet at each table.

Millisecond recognition and settlement

The moment cards settle they are read, the game's rules are applied to points, draws and outcome, and the result is out within 300 milliseconds.

Dealer verification on a tablet

Every result appears on the tablet at the table. If a read looks wrong the dealer corrects it on the spot, and the correction is logged in full.

One server for the whole floor

A single ingest server collects every table's IP camera feed and pushes it to the platform. No compute box per table.

An image kept for every card

Each round stores card images, timestamps and confidence scores. Disputes are answered from one lookup instead of a footage hunt.

No table modification

Mount an overhead IP camera on a bracket, or reuse the camera already there. A table goes live in hours, without closing it.

Open API integration

REST, WebSocket and Webhook outputs feed roadmap boards, live-dealer platforms, lobby displays, BI reporting and your existing systems.

Both shoe types supported

A plain transparent shoe or an AngelEye smart shoe you have already deployed — either works, and both can run at once so each verifies the other.

Transparent shoe
AI vision recognition

Transparent shoe

Keep the transparent shoe you already use — no electronic shoe required. An overhead IP camera captures the felt and AI vision reads every card.

  • No shoe replacement, lowest cost to deploy
  • Works for any table game using standard playing cards
  • The card image doubles as your dispute evidence
AngelEye smart shoe SMART SHOE
Shoe data + vision, two sources

AngelEye smart shoe

If your floor already runs AngelEye smart shoes, plug straight in. The shoe emits every drawn card over RS-232, converted to Ethernet by a Moxa serial device server and fed to the ingest server.

  • AngelEye Ex II
  • Super AngelEye
  • Supports AngelEye Ex II and Super AngelEye
  • Moxa NPort serial-to-Ethernet, no change to the shoe
  • Run it alongside vision so the two sources verify each other

AngelEye is a trademark of Angel Group Co., Ltd. Moxa and NPort are trademarks of Moxa Inc. This page describes integration compatibility only and implies no partnership or endorsement.

How it works

Four stages. Your floor owns the hardware in the first two; we deliver the rest.

  1. 01

    IP camera capture

    An overhead IP camera holds a fixed view of the card area, covering every betting box and third-card position.

  2. 02

    Ingest and push

    One ingest server on site collects every table's camera feed, encodes it and pushes it to the streaming platform.

  3. 03

    AI recognition and settlement

    The platform detects each card, reads rank and suit, then applies the game's rules for draws, outcome and side bets.

  4. 04

    Tablet check and output

    The result lands on the dealer tablet for verification, then goes out to roadmap boards, displays and your back office.

Learn more

Card games supported

Recognition is decoupled from game rules: the core identifies cards, a rules module settles the round. Any playing-card game can therefore be supported — only the order of development differs.

See the full game list

Ways to deploy

Choose against what your floor already has. Nothing here requires rebuilding tables or re-architecting your network.

Full hardware package

IP cameras, brackets, PoE switching, ingest server and dealer tablets delivered together — right for new tables or for the best possible capture conditions.

Reuse existing cameras

Keep the overhead IP cameras you already run and add only the ingest server and tablets. Lowest cost and lowest install effort.

Single-table pilot

Run one table live, measure real accuracy, latency and workflow impact on your floor, then scale once you are satisfied.

Want to see it running on your own table?

Send us photos or footage of your table and we will run an offline assessment, then report the accuracy and latency actually achievable in your conditions.