FocusGuard (OpenCV + Python)

FocusGuard is a real-time webcam monitor for this specific rule:

• Detect a phone in the camera view.
• Detect your face.
• If you appear to be looking at that phone continuously for more than 4 seconds, play a beep warning.

How It Works

The app now uses a phone-first pipeline.

Detection layer:

• Face, eyes, smile, lips: OpenCV Haar/lip heuristics.
• Phone: YOLOv4-tiny (COCO class: cell phone) through OpenCV DNN.

Phone-looking layer:

• Select primary face (largest face box).
• Select primary phone (highest-confidence phone box).
• Consider user "looking at phone" when phone is below the face and horizontally close to it.
• If this condition continues for threshold seconds (default 4.0s), fire a beep alert.

Note: this is still a heuristic for "looking at phone" from webcam geometry.

Project Structure

FocusGuard/
  README.md
  requirements.txt
  .gitignore
  src/
    app.py
    attention/
      __init__.py
      screen_attention_monitor.py
    detectors/
      __init__.py
      haar_feature_detector.py
      phone_detector.py
    utils/
      __init__.py
      drawing.py

Setup

1. Install dependencies with system Python:

py -3.13 -m pip install -r requirements.txt

On first run, model files are downloaded automatically to models/yolo/.

Run

Basic run:

py -3.13 -m src.app

Custom camera and thresholds:

py -3.13 -m src.app --camera 0 --width 1280 --height 720 --phone-threshold 4.0 --phone-confidence 0.35

Runtime Display

• Bounding boxes for face, eyes, smile, and lips.
• Phone timer overlay, for example: Phone timer: 2.8s / 4.0s.
• State text, for example: Looking at phone, No phone, No face, Phone too far from face.
• Alarm when threshold is crossed.

Tunable Parameters

• --phone-threshold: seconds before warning, default 4.0.
• --phone-confidence: detector confidence threshold, default 0.35.

Platform Note

• On Windows, alarm uses winsound.Beep.
• If beep API is unavailable, the app falls back to terminal bell/message.

Controls

• Press q to quit.

Practical Tips

• Use good front lighting for stable face/eye detection.
• Keep camera at eye level.
• Increase --phone-threshold if alerts fire too quickly.
• Reduce --phone-confidence slightly if phone is missed often.