SafeWatch - Women Safety Analytics using ML (DeepFace)

Project Overview

This project is designed to enhance women's safety using advanced Machine Learning (ML) models for gender bias detection. It integrates facial recognition, gesture recognition, and audio analysis to monitor environments in real-time and proactively alert in cases of potential danger or abnormal gender ratios. By analyzing video feeds frame-by-frame and detecting certain gestures like SOS signals, this system aims to provide early warnings and help prevent in unsafe situations.

---

Features

• Real-Time Gender Bias Detection: Analyze video and audio streams to identify gender imbalances in a scene, particularly in sensitive environments.

• Gesture Recognition for SOS Alerts: Detect SOS or similar distress gestures using hand gesture recognition powered by MediaPipe.

• Proactive Alerting: Alerts are triggered based on scenarios such as a high male-to-female ratio, sudden increases in male presence, or large group gatherings.

• Voice and Gesture Integration: The system integrates multiple modalities (voice, face, gestures) for a comprehensive safety solution.

• Privacy-Focused AI: Built with privacy and ethical considerations, this solution processes data securely without retaining unnecessary information.

---

Project Structure

├── models/                     # Contains ML models for face and gender detection
├── src/
│   ├── video_analysis.py        # Video processing logic, breaking video into frames for analysis
│   ├── gender_bias_detection.py # DeepFace-based gender bias detection
│   ├── gesture_detection.py     # Gesture detection using MediaPipe
│   └── audio_analysis.py        # (Optional) Audio processing for speech recognition
├── data/                        # (Optional) Placeholder for sample videos or images
├── requirements.txt             # Python dependencies
└── README.md                    # Project documentation

---

Key Components

Video Analysis

This component processes video feeds or webcam streams, breaking them into frames to detect the number of men and women using DeepFace.

Gender Bias Detection

Using DeepFace, the system determines the gender of individuals in the frame and compares gender ratios. If suspicious patterns (e.g., a 4:1 male-to-female ratio) are detected, an alert is generated.

Gesture Detection

Powered by MediaPipe, the system analyzes hand gestures to detect SOS or distress gestures. If an SOS gesture is detected, the system sends a real-time alert.

---

Alerts

Alerts are generated based on predefined suspicious activity scenarios:

• High male-to-female ratio (e.g., 4:1 or higher)
• Large group gatherings (more than 10 people in the frame)
• Sudden changes in gender counts
• SOS gestures detected via hand movements

---

Setup Instructions

Prerequisites

• Python 3.7 or above
• OpenCV
• MediaPipe
• DeepFace
• NumPy
• (Optional) Google Colab for cloud-based execution

---

pip install -r requirements.txt

Running the Program:

Webcam-based Detection:

python src/webcam_analysis.py

Video-based Detection:

Update the video path in the script and run:

python src/video_analysis.py --video /path/to/video.mp4

---

How It Works

• Gender Detection: The DeepFace model analyzes each frame of the video or live feed to detect the gender of each person in the frame. If the male-to-female ratio becomes imbalanced (e.g., 4 men to 1 woman), an alert is triggered.

• Gesture Recognition: Using MediaPipe, the system monitors hand gestures in real-time. If an SOS gesture (or similar) is detected, it flags the frame as suspicious.

• Alerting Mechanism: Alerts are triggered when predefined conditions (e.g., large groups, high male-to-female ratios, SOS gestures) are detected.

---

Scenarios Detected

• High Male-to-Female Ratio: Triggered when the male population in a frame is disproportionately higher than the female population.
• Large Groups: Detects gatherings of more than 10 people in a frame.
• SOS Gesture: Detects distress signals through hand gestures using MediaPipe.
• Sudden Changes in Gender Count: Flags frames where the count of men or women changes drastically from one frame to another.

---

Customization

You can customize the alert scenarios in gender_bias_detection.py and gesture_detection.py by adjusting the parameters for ratios, group sizes, or specific gesture conditions.

---

Future Enhancements

• Audio Analysis: Integrate speech recognition and voice tone analysis to detect distress calls.
• Mobile App Integration: Provide alerts through a mobile app for immediate notifications.
• Crowd Density Monitoring: Enhance the system to analyze crowd density in large spaces.

---

Contributing

Feel free to contribute to this project by opening a pull request or raising an issue on GitHub. We welcome suggestions for new features, optimizations, or bug fixes.

---

License

This project is licensed under the MIT License. See the LICENSE file for more details.

---