F2: Real-Time Optical Frame Loop & Debounce Heuristics

1. Description & Debounce Logic

Inside ChefVideoPlayer.jsx, an active requestAnimationFrame loop continuously checks if a new video frame is ready (readyState === 4). Detected gestures are filtered through two safety gates:

1. Confidence Threshold: $Score > 0.75$ (rejects ambiguous hand poses).
2. Time-Decay Debounce: $Date.now() - lastActionTime > 1500ms$ (prevents rapid-fire toggling).

2. Heuristic Filter Flow

flowchart TD
    Frame[New Optical Video Frame] --> Ready{readyState == 4?}
    Ready -->|No| NextTick[Skip to Next RAF]
    Ready -->|Yes| Recognize[recognizeVideo Frame]
    Recognize --> ConfCheck{Confidence > 0.75?}
    ConfCheck -->|No| Discard[Discard Ambiguous Pose]
    ConfCheck -->|Yes| TimeCheck{Elapsed Time > 1500ms?}
    TimeCheck -->|No| Throttle[Suppress: Debounce Active]
    TimeCheck -->|Yes| FireAction[Execute Video Action & Reset Timer]

3. Say It Out Loud

"One of the biggest challenges in optical gesture control is preventing accidental double-triggers, known as gesture flapping. I solved this by implementing two distinct heuristic gates. First, any gesture classification with less than 75% confidence is discarded immediately. Second, I introduced a 1500ms time-decay debounce window stored in a React ref. Once a command like Pause fires, the system temporarily suppresses subsequent gesture executions for 1.5 seconds, giving the user ample time to lower their hand naturally without accidentally un-pausing the video."