FocusCam

FocusCam is a comprehensive, real-time attention and distraction monitoring system built with Python. It leverages your webcam to analyze facial cues and environmental objects to determine if a user is focused, drowsy, or distracted. The system combines multiple AI models to provide a robust and nuanced assessment of user attention.

Features

• Hybrid Gaze Tracking: Utilizes a powerful combination of MediaPipe's high-fidelity iris tracking for rule-based analysis and a custom-trained TensorFlow model for deep learning-based gaze direction classification (forward, up, down, left, right).
• Yawn Detection: Employs a custom CNN model, trained on the YawDD dataset, to specifically detect yawns as an indicator of drowsiness.
• Phone Detection: Integrates a YOLOv8 model to identify when a user is looking at a cell phone, a primary source of distraction.
• Combined Attention Score: Prioritizes and aggregates statuses from all detectors (phone > yawn > gaze) to provide a single, debounced, and easy-to-understand status: "PAYING ATTENTION" or "DISTRACTED".
• Real-Time Visual Feedback: Overlays status, diagnostic information, and targeting guides directly onto the webcam feed for immediate user feedback.

How It Works

The core logic resides in combined.py, which establishes a multi-layered detection pipeline that runs in real-time on a webcam feed.

1. Frame Capture: The system captures frames from the default webcam.
2. Face & Landmark Detection: MediaPipe's Face Mesh is used to detect the face and its 478 landmarks, including precise iris, eye, and mouth coordinates.
3. Distraction Analysis (Prioritized):
   • Phone Detection: The frame is first passed to a YOLOv8 model. If a "cell phone" is detected with sufficient confidence, the status is immediately set to "DISTRACTED: PHONE".
   • Yawn Detection: If no phone is detected, the mouth region is cropped based on facial landmarks and fed into a custom-trained yawn detection model. If a yawn is detected, the status is set to "DISTRACTED: YAWNING".
   • Gaze Detection: If neither a phone nor a yawn is detected, the system analyzes the user's gaze. It calculates the relative position of the irises within the eyes. This rule-based vector is combined with a prediction from a custom gaze classification model. If the final verdict is not "forward", the status is set to "DISTRACTED: LOOKING AWAY".
4. Status Display: If none of the above distraction conditions are met, the status is "PAYING ATTENTION". The final status is debounced to prevent flickering and displayed clearly on the screen.

Models

This repository includes several pre-trained models.

• Gaze Model (gaze_model_1.keras): A Keras CNN model trained on a custom dataset (eye_dataset/) of cropped eye images, created using the gazefarm.py script. The training process is documented in GazeTester.ipynb.
• Yawn Model (yawn_model.h5): A Keras CNN based on MobileNetV2, trained to classify mouth crops as "yawn" or "no_yawn". The data preprocessing and training pipeline, which uses the YawDD video dataset, is detailed in NewYawn.ipynb.
• Object Detection Model (yolov8n.pt): The standard YOLOv8-nano model from Ultralytics, used for its efficiency and accuracy in detecting common objects, including cell phones.

Installation

To get started with FocusCam, follow these steps.

Prerequisites:

• Python 3.8+
• A connected webcam

Setup:

1. Clone the repository to your local machine:

   git clone https://github.com/executioner501/FocusCam.git
   cd FocusCam

2. Install the required Python packages using the requirements.txt file:

   pip install -r requirements.txt

Usage

To run the main application, execute the combined.py script.

python combined.py

A window will open showing your webcam feed with the attention status and diagnostic information overlaid. Press the ESC key to close the application.

Individual Testers

You can test each component of the system individually:

• Gaze Detection: python newgaze.py
• Yawn Detection: python yawn_tester.py
• Phone Detection: python phone_tester.py

Data Collection

To collect your own eye images for retraining the gaze model, you can use the gazefarm.py script.

python gazefarm.py

Press f, l, r, u, d to label the direction you are looking (forward, left, right, up, down) and the script will automatically save cropped eye images to the eye_dataset directory.

Contributors