AI Table Watch β€” Camera Module
TAVORA Β· PHASE 2 PILOT MODULE
SIMULATED PREVIEW β€” NO REAL CAMERA
KILL SWITCH

FLOOR β€” SCHEMATIC TOP-DOWN VIEW (NOT CAMERA IMAGERY)

Demo snapshot cadence is accelerated (4s). The real module classifies a still snapshot every 15–30s, and a state only changes after it persists across 3 consecutive snapshots β€” the β€œcandidate” counter under each table shows that rule working.

⚠ ATTENTION ALERTS

No tables need attention right now.

SNAPSHOT LOG β€” EVERY FRAME CLASSIFIED, THEN DELETED

PRIVACY β€” HOW THIS MODULE WORKS (CAMERA PRIVACY PACK)

  • Snapshots, never video. One still frame every 15–30 seconds. No live feed, no recording, no playback β€” for anyone.
  • Frames are deleted seconds after classification. Only the state label + confidence leave the edge box. This log proves it: every entry ends β€œframe deleted β€” 0 stored”.
  • No faces, no biometrics, no audio. The classifier reads table scenes (plates, glasses, occupied chairs) β€” never who anyone is.
  • Pilot tables only, privacy-masked to the table area; neighbouring tables, bar, doors and windows are masked out.
  • 3-snapshot rule: one odd frame never flags a table β€” a state must persist across 3 consecutive snapshots.
  • Kill switch: one tap (top right) halts all snapshots instantly. Off means off.
  • Built for BC PIPA. Activation requires the venue's signed privacy addendum (Owner Dashboard β†’ Settings) plus the one-time BC privacy-lawyer review before a paying rollout.