🛡️ FaceShield — NHAI Hackathon 7.0

Offline Facial Recognition & Liveness Detection System for Remote Field Operations

A fully offline, AI-powered biometric authentication module for the Datalake 3.0 React Native app — enabling secure field personnel attendance in zero-network environments.

---

📋 Table of Contents

• Overview
• Features
• Technical Specs
• Architecture
• Installation
• Usage
• API Reference
• Performance Benchmarks
• Project Structure
• Team

---

Overview

FaceShield solves the critical problem of biometric authentication for NHAI field personnel who operate in remote areas with zero internet connectivity. The system integrates as a React Native native module into the existing Datalake 3.0 app.

Problem: Field workers in highway maintenance zones cannot be authenticated biometrically without internet — allowing proxy fraud and manual attendance manipulation.

Solution: A lightweight (<20 MB) offline AI module combining:

• MobileFaceNet (TFLite) for face recognition
• MobileNetV3 + MediaPipe for dual-layer liveness detection
• AES-256 encrypted SQLite for offline storage
• AWS S3 sync + purge when connectivity is restored

---

Features

• ✅ 100% Offline — no internet required for authentication
• ✅ Dual-Layer Liveness — active challenge (blink/smile/turn) + passive texture anti-spoofing
• ✅ ~8 MB total model bundle (2.5× under the 20 MB limit)
• ✅ <650ms end-to-end authentication on mid-range devices
• ✅ >99% accuracy with Indian demographic fine-tuning
• ✅ AES-256 encryption for all local biometric data
• ✅ AWS sync & purge on network restoration
• ✅ React Native — works on Android 8.0+ and iOS 12+
• ✅ Apache 2.0 — fully open source, zero license fees

---

Technical Specs

Parameter	Requirement	FaceShield
Framework	React Native (Android + iOS)	✅ RN 0.73+
Model Size	~20 MB target	✅ ~8 MB total
Speed	< 1 second	✅ ~650ms
Min OS	Android 8.0+ / iOS 12+	✅ Supported
Min RAM	3 GB	✅ Optimized
Accuracy	> 95%	✅ ~99%
Open Source	Yes	✅ Apache 2.0
AWS Sync	Yes	✅ + Purge logic

---

Architecture

┌─────────────────────────────────────────────────────────┐
│                   Datalake 3.0 App                      │
│  ┌──────────────────────────────────────────────────┐   │
│  │              FaceShield Module                   │   │
│  │                                                  │   │
│  │  Camera → FaceDetection → LivenessCheck          │   │
│  │               ↓                ↓                 │   │
│  │         MobileFaceNet    MobileNetV3              │   │
│  │         (TFLite 2MB)   + MediaPipe               │   │
│  │               ↓                                  │   │
│  │         SQLite (AES-256 encrypted)               │   │
│  │               ↓ (on network restore)             │   │
│  │         AWS S3 Sync → Lambda ACK → Purge         │   │
│  └──────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────┘

See docs/architecture.md for full technical architecture.

---

Installation

Prerequisites

• Node.js 18+
• React Native CLI
• Android Studio / Xcode
• Java 17 (for Android)

Setup

# Clone the repository
git clone https://github.com/Ani-sha23/hackathon7-faceshield.git
cd hackathon7-faceshield

# Install dependencies
npm install

# iOS (Mac only)
cd ios && pod install && cd ..

# Android
npx react-native run-android

# iOS
npx react-native run-ios

TFLite Models

The TFLite model files are stored in:

• Android: android/app/src/main/assets/models/
• iOS: ios/FaceShield/models/

Models included:

• mobilefacenet.tflite — face recognition (~2 MB)
• antispoofing.tflite — passive liveness detection (~3 MB)

---

Usage

Basic Authentication Flow

import FaceShield from './src/services/FaceShieldService';

// 1. Enroll a user (requires network for initial setup)
await FaceShield.enrollUser({
  userId: 'EMP001',
  photos: [photo1, photo2, photo3, photo4, photo5], // 5 photos
});

// 2. Authenticate (fully offline)
const result = await FaceShield.authenticate({
  userId: 'EMP001',
  onChallenge: (challenge) => {
    // challenge = 'BLINK' | 'SMILE' | 'TURN_LEFT' | 'TURN_RIGHT'
    console.log('Perform:', challenge);
  },
});

if (result.authenticated) {
  console.log('✅ Verified:', result.confidence);
  // Mark attendance
} else {
  console.log('❌ Failed:', result.reason);
}

Sync Attendance Records

import SyncService from './src/services/SyncService';

// Manually trigger sync (auto-triggers on network restore)
await SyncService.syncPendingRecords();

---

API Reference

See docs/api-reference.md for full API documentation.

Core Methods

Method	Description
FaceShield.enrollUser(params)	Enroll a new user with 5 face photos
FaceShield.authenticate(params)	Authenticate with liveness check
FaceShield.deleteEnrollment(userId)	Remove user enrollment
SyncService.syncPendingRecords()	Upload pending attendance to AWS
StorageService.getPendingCount()	Get count of unsynced records

---

Performance Benchmarks

Device Tier	Device Example	Auth Time	RAM Usage
Low-end	Redmi 9A (3GB, SD460)	~920ms	~110 MB
Mid-range	Redmi Note 11 (4GB, SD680)	~650ms	~95 MB
High-end	Samsung A53 (6GB, SD778G)	~380ms	~82 MB

All devices tested on Android 12 with ambient daylight conditions.

---

Project Structure

hackathon7-faceshield/
├── src/
│   ├── components/
│   │   ├── CameraView.js          # Live camera feed component
│   │   ├── LivenessChallenge.js   # Active challenge UI
│   │   └── ResultOverlay.js       # Auth result display
│   ├── screens/
│   │   ├── EnrollmentScreen.js    # User enrollment flow
│   │   ├── AuthScreen.js          # Authentication screen
│   │   └── AttendanceScreen.js    # Attendance log view
│   ├── services/
│   │   ├── FaceShieldService.js   # Main auth orchestrator
│   │   ├── FaceRecognition.js     # MobileFaceNet TFLite wrapper
│   │   ├── LivenessService.js     # Liveness detection logic
│   │   ├── StorageService.js      # SQLite + AES encryption
│   │   └── SyncService.js         # AWS S3 sync & purge
│   ├── models/
│   │   └── ModelLoader.js         # TFLite model initialization
│   └── utils/
│       ├── ImageUtils.js          # Preprocessing helpers
│       └── CryptoUtils.js         # AES-256 encryption utils
├── docs/
│   ├── architecture.md            # Technical architecture
│   ├── api-reference.md           # API documentation
│   ├── integration-guide.md       # Datalake 3.0 integration
│   └── benchmarks.md              # Performance benchmarks
├── android/
│   └── app/src/main/assets/models/  # TFLite model files
├── ios/
│   └── FaceShield/models/           # TFLite model files
├── LICENSE
└── README.md

---

Team

Name	Role
Anisha Garg	Team Leader — AI/ML, React Native
Team Member 1	Backend, AWS Integration

Hackathon: NHAI Hackathon 7.0 Submission Date: 01.06.2026

---

License

This project is licensed under the Apache License 2.0 — see the LICENSE file for details.

All third-party components used are open source (Apache 2.0 / MIT). No proprietary or licensed software is used.

---

Built with ❤️ for NHAI field personnel safety