feat(ai): modified blur detector by adding edge density and flicker rate

[DhanviNautiyal]

Implement real-time detection of virtual background / software blur using edge AI in the browser.

Previously, blur detection relied only on Laplacian variance, which was insufficient for detecting virtual backgrounds or software-applied blur.

This update enhances detection by combining:

• Laplacian variance (blur detection)
• Sobel edge density (edge loss)
• Background variance using segmentation mask
• Frame-to-frame flicker detection

The detection runs fully on the client (Edge AI), ensuring privacy and real-time performance.

Key Changes

Frontend:

• Enhanced BlurDetectorWorker.ts with:
  • Sobel operator for edge density
  • Background variance calculation
  • Weighted anomaly scoring
• Updated BlurDetector.tsx:
  • Integrated MediaPipe Selfie Segmentation
  • Added throttling to anomaly reporting (5s interval)

Backend:

• Reused existing BLUR_DETECTION anomaly pipeline
• No schema changes required

Improvements

• Detects virtual blur
• Reduces false positives
• Works in real-time
• No backend video processing required

---

Dependencies

• depends on: none