🛡️ SafeCity — AI-Powered Personal Safety Web App

Status: Core Features Implemented — Backend, Sensor Integration, and Alert System Prototyped

---

🚀 Overview

SafeCity is an intelligent web-based safety monitoring application designed to detect real-time distress through audio, visual, motion, and contextual signals.
Using modern web technologies and AI inference models, the system provides tiered safety alerts that escalate automatically when signs of distress are detected.

This half submission focuses on core functional integration (AI models, alert logic, and backend connectivity) with minimal UI and stable background operation.

---

🧩 Core Modules Implemented

1. Real-Time Audio Analysis

• Captures live microphone input using the Web Audio API.
• Runs TensorFlow.js Wav2Vec2 model for scream and panic detection.
• Operates with 3-second inference windows.
• Outputs distress probability to Alert Engine.

2. Vision Distress Detection

• Uses MediaPipe FaceMesh and BlazePose.
• Extracts facial emotion and posture indicators.
• Detects visual stress patterns (e.g., raised eyebrows, abrupt motion).
• Streams scores to the AI Alert Engine.

3. Contextual Risk Assessment

• Implements browser Geolocation API and local time awareness.
• Dynamically adjusts risk based on:
  • Time of day (higher weight at night)
  • Risk zones (configurable data source)
• Provides normalized risk score (0–1).

4. AI Alert Engine

• Fusion algorithm integrating multimodal inputs: total_risk = (0.4 * audio) + (0.3 * vision) + (0.2 * motion) + (0.1 * context)

• Triggers alerts in progressive tiers:

• ⚠️ Level 1: User confirmation prompt (“Are you okay?”)

• ⚡ Level 2: Device vibration alert

• 🚨 Level 3: SOS message via Twilio

5. Emergency Contact Integration (Twilio)

• FastAPI backend routes triggered from frontend.
• Sends live SMS alerts containing:
• GPS coordinates
• Timestamp
• “SafeCity Alert” message
• Secure credential handling via environment variables.

6. Background Operation Mode

• Implemented as a PWA (Progressive Web App).
• Uses Service Workers and Background Sync for persistence.
• Includes a minimal “Safety Status Bar” UI.

---

🧠 Tech Stack

Layer	Technology	Description
Frontend	React + TensorFlow.js + MediaPipe	UI + AI processing
Backend	FastAPI + Twilio	Alert routing + SMS integration
Hosting	Vercel (Frontend) + Render/AWS (Backend)	Deployment
Database	PostgreSQL (Optional)	Logging alerts and metadata
AI Models	Wav2Vec2, BlazePose, FaceMesh	Multimodal inference

---

🎨 UI / UX Specifications

• Primary Color: #2E3B55
• Background Color: #1A2133
• Accent Color: #465A80
• Font Pairing:
• Space Grotesk — Headlines
• Inter — Body text
• Design Philosophy:
• Minimal, always-on interface
• Subtle animations
• High-contrast indicators for safety state

---

⚙️ Installation (Local Setup)

Prerequisites

• Node.js ≥ 18.x
• Python ≥ 3.10
• Twilio Account with API Credentials
• Secure HTTPS environment (required for sensors)

Steps

1. Clone Repository

git clone https://github.com/<your-username>/SafeCity.git
cd SafeCity

2. Install Frontend

cd frontend
npm install
npm run dev

3. Install Backend

cd ../backend
pip install -r requirements.txt
uvicorn main:app --reload

4. Configure Environment Variables

Create .env file in /backend:

TWILIO_ACCOUNT_SID=your_sid_here
TWILIO_AUTH_TOKEN=your_auth_token_here
TWILIO_PHONE_NUMBER=+1234567890

---

🧪 Current Limitations

• Emotion detection optimized for daylight; poor accuracy in low light.
• Audio model lacks multi-language distress detection.
• Background motion sensing partially implemented.
• No user account or contact management UI yet.
• Battery optimization pending for continuous monitoring.

---

🧾 License

MIT License © 2025 SafeCity Developers Team

---

👥 Contributors

• AI & Vision Systems: Sousnigdho Das
• Backend & Infrastructure: Sousnigdho Das
• UI/UX Designer: Shreeya Banerjee
• Tester: Pranav Sharma
• App Developer: Cynthia Pachal

SafeCity — Empowering safety through real-time AI.