This repository is archived.
It is kept as a personal learning and experimentation project and is no longer actively developed.

AWS-Based Smart Attendance System using Face Recognition (IoT + Cloud)

📦 This repository contains an academic, proof-of-concept project developed as part of a college course.
It demonstrates an end-to-end IoT → Cloud → AI pipeline for automated attendance marking using face recognition.

The repository is intended as a learning and architectural showcase and is not actively maintained.

⚠️ Important Disclaimer (Read First)

Warning

Educational & Ethical Use Only

• This project was developed strictly for academic purposes.
• All face recognition was conducted in a consensual, controlled environment.
• The system is a prototype and not intended for production deployment.
• No real credentials, private keys, or secrets are included in this repository.
• Any sensitive values (Wi-Fi credentials, AWS certificates, endpoints) have been intentionally redacted or replaced with placeholders.

Use responsibly and in compliance with privacy, biometric, and data-protection regulations.

📖 Introduction

Background & Motivation

Manual attendance systems are inefficient, error-prone, and difficult to scale—especially in environments such as classrooms and laboratories. With the growth of IoT devices, cloud computing, and AI services, it is now possible to automate such workflows in a reliable and scalable manner.

This project explores how a low-cost edge device (ESP32-CAM) can be integrated with AWS cloud services to create a fully automated attendance pipeline powered by face recognition.

The focus of the project is not just face recognition, but the complete system design:

• Edge data capture
• Secure device-to-cloud communication
• Event-driven cloud processing
• Serverless architecture
• Persistent attendance storage

🎯 Objectives

• Build an end-to-end attendance system using IoT and AWS
• Capture images periodically from an ESP32-CAM
• Transmit images securely using MQTT over TLS
• Process images using AWS Lambda
• Perform face recognition using AWS Rekognition
• Store attendance records in DynamoDB
• Demonstrate real-world use of event-driven serverless architecture

🧠 High-Level System Overview

What the system does

1. ESP32-CAM captures an image every 10 seconds
2. Image is encoded as Base64
3. Image is sent securely to AWS IoT Core via MQTT
4. AWS IoT Rule triggers a Lambda function
5. Image is decoded and stored in Amazon S3
6. S3 upload triggers:
   • Face indexing (for enrollment)
   • Face matching (for attendance)
7. Attendance is marked in DynamoDB based on:
   • Face match confidence
   • Time difference (≥ 30 minutes)
   • Class period (IST)

🏗️ Architecture Diagram

flowchart LR
    A[ESP32-CAM] -->|MQTT over TLS| B[AWS IoT Core]
    B -->|IoT Rule| C[Lambda: Image Decoder]
    C -->|JPEG Upload| D[S3 Bucket]
    D -->|S3 Event| E[Lambda: Face Indexing]
    D -->|S3 Event| F[Lambda: Face Matching]
    E -->|Index Faces| G[AWS Rekognition Collection]
    F -->|Search Faces| G
    F -->|Attendance Update| H[DynamoDB]

Loading

🔌 Edge Device: ESP32-CAM

Responsibilities

• Capture images using onboard camera
• Encode image data to Base64
• Publish JSON payloads to AWS IoT Core
• Maintain secure TLS connection using X.509 certificates

Key Characteristics

• Image capture interval: 10 seconds
• Communication protocol: MQTT
• Transport security: TLS (Port 8883)

🔒 Note: All credentials, certificates, and private keys are intentionally excluded from this repository.

☁️ Cloud Components (AWS)

1️⃣ AWS IoT Core

• Acts as the secure entry point for IoT messages
• Authenticates ESP32 using X.509 certificates
• Routes messages using IoT Rules

2️⃣ Lambda — Image Ingestion

Trigger: AWS IoT Rule Function:

• Decodes Base64 image payload
• Generates timestamp-based filename
• Uploads image to Amazon S3

3️⃣ Amazon S3

• Temporary storage for captured images
• Acts as an event source for downstream processing

4️⃣ Lambda — Face Indexing

Trigger: S3 Object Created Function:

• Indexes detected faces into a Rekognition Collection
• Uses ExternalImageId (mapped to student roll number)

5️⃣ Lambda — Attendance Processing

Trigger: S3 Object Created Function:

• Searches face against Rekognition Collection
• Applies confidence threshold (≥ 95%)
• Converts timestamp to IST
• Determines class period
• Updates attendance in DynamoDB

6️⃣ AWS Rekognition

• Managed face recognition service
• Stores facial embeddings in a persistent collection
• Used as-is, without custom ML logic

7️⃣ DynamoDB

• Stores attendance records using:

  • RollNumber (Partition Key)
  • Date (Sort Key)

• Tracks:

  • In-time
  • Out-time
  • Attendance status per period

⏱️ Attendance Logic

• Attendance periods: 9 AM – 4 PM (IST)

• Periods are hourly (1–7)

• A student is marked Present if:

  • Face match confidence ≥ 95%
  • Time difference between first and last detection ≥ 30 minutes (to reduce misuse and ensure sustained presence)
  • Detection occurs within a valid period

📂 Project Structure (Logical)

.
├── esp32/
│   └── esp32_cam_mqtt.ino
│
├── lambdas/
│   ├── image_ingest_lambda.py
│   ├── face_index_lambda.py
│   └── attendance_lambda.py
│
├── README.md
└── .gitignore

🔐 Security Considerations

• No secrets are committed

• All credentials are replaced with placeholders

• .gitignore excludes:

  • Certificates
  • Private keys
  • Environment configs

• Designed following least-privilege IAM principles

⚖️ Ethics & Privacy

• Explicit consent was assumed during data collection
• System was deployed in a controlled academic environment
• No public or unauthorized biometric data was processed
• This repository exists purely for educational demonstration

🧪 Project Status

• ✅ Core pipeline implemented
• ❌ Frontend / dashboard not implemented
• ❌ Firebase integration planned but not completed
• 📦 Repository archived as a learning artifact

🏁 Conclusion

This project demonstrates how IoT, cloud services, and AI can be combined to build a real-world, event-driven system using modern architectural patterns.

Beyond face recognition, the project highlights:

• Secure device-to-cloud communication
• Serverless, scalable processing
• Practical use of managed AI services
• Clean separation between edge, compute, and data layers

It serves as a strong academic and architectural showcase for cloud-native system design.

📜 License

MIT License Provided as-is, for educational and learning purposes.

This repository is archived.
It is kept as a personal learning and experimentation project and is no longer actively developed.