Computer Vision · Live Table

Reading the whole table,
one deal at a time.

A real-time system that watches a live blackjack table with a single camera, recognizes every card, scores each hand against the dealer, and lights up the winners on a physical LED strip.

52-class card model 0.95 mAP@50 7 seats + dealer ~instant LED feedback
Live card detection on a blackjack table

Live detection frame — every card boxed with its rank, suit and confidence, LEDs glowing under the felt.

How it works

One phone camera in, a lit-up winning seat out. Five stages run every round.

01
📷

Capture

A phone camera (iVCam) feeds a top-down view of the felt.

02
🧠

Detect

Tiled YOLOv8 finds all 52 card classes, even tiny corner indices.

03
🎯

Group

Cards snap to 7 fixed seats on an arc, plus the dealer line.

04
🃏

Score

Each hand is totalled and compared to the dealer — win, lose, push, blackjack.

05
💡

Light

A Raspberry Pi LED strip lights winning seats in their seat color, near-instantly.

Fixed-seat table geometry

Cards are assigned to seats by geometry, so the LED mapping stays rock-solid no matter how the cards land.

Seat layout geometry overlaid on the table
  • 7 player seats sit on an ellipse; the dealer owns everything below the red line.
  • Every card routes to its nearest seat point where the seat's center line meets the arc.
  • Geometry is fully tunable — apex, arc angle, radius and dealer line are all constants.
  • Each seat carries a fixed color, shared by the on-screen boxes and the physical LEDs.
1 2 3 4 5 6 7

Tracked across real play

The system keeps running game stats for every seat and the dealer over live sessions.

139
Rounds dealt
760
Player hands
36
Blackjacks
44
Dealer busts

Under the hood

Trained on ~42 real corrected photos over a 24k-image base, fine-tuned at 1280px so the tiny corner indices survive.

PythonYOLOv8 / UltralyticsOpenCV Tiled inferenceRoboflowRaspberry Pi SPI LED stripPersistent SSH bridgeiVCam