AWS Face Recognition (Serverless & Auto-scaling)

A comprehensive AWS-based face recognition system that demonstrates:

• Video Splitting with AWS Lambda (using FFmpeg)
• Face Detection & Recognition with a pre-trained CNN (ResNet)
• Auto-scaling on EC2 based on SQS message queue depth
• Serverless Pipelines (S3 triggers, asynchronous Lambda invocations)

Note: This project contains subfolders for different parts (Project1, Project2, Project3).

Table of Contents

1. Overview
2. Repository Structure
3. Architecture
4. Prerequisites
5. Setup & Installation
6. Usage
7. Security Considerations
8. Possible Improvements
9. Acknowledgments
10. License
11. Contact

Overview

This repository showcases an end-to-end face recognition pipeline using AWS:

• Input Bucket: Users upload .mp4 videos to <ASU_ID>-input.
• Video-Splitting Lambda: Extracts frames from uploaded videos and stores them in <ASU_ID>-stage-1.
• Face-Recognition Lambda: Detects faces, classifies them, and writes the result to <ASU_ID>-output.
• Auto-scaling App Tier: Launches or terminates EC2 instances based on SQS queue length.
• Web Tier: A Spring Boot REST API that interacts with SQS and S3.

Originally developed for a Cloud Computing course, but it’s a solid reference for real-world AWS patterns.

Repository Structure

aws-face-recognition/
├── Project1/                        # Basic AWS resource management (EC2, S3, SQS)
├── Project2/                        # IaaS-based face recognition with auto-scaling
│   ├── src/
│   ├── pom.xml
│   └── ...
├── Project3/                        # PaaS-based (Lambda) video splitting & face recognition
│   ├── video-splitting/
│   ├── face-recognition/
│   └── ...
├── .gitignore
├── README.md                        # You're here!
└── Other scripts/data as needed

Architecture

A high-level flow:
User (video) -> [S3: <ASU_ID>-input] -> [Lambda: video-splitting] -> [S3: <ASU_ID>-stage-1] -> [Lambda: face-recognition] -> [S3: <ASU_ID>-output]

For the IaaS version (Project 2):
User -> Web Tier (Spring Boot) -> SQS -> [Auto-Scaling EC2 App Tier] -> S3

Each part uses AWS components:

• EC2 for the App Tier.
• S3 for input/output storage.
• SQS for message queue handling.
• Lambda for serverless tasks (video splitting & face recognition).
• IAM for access control (best done with roles rather than keys in code).

Prerequisites

1. AWS Account
   • S3, EC2, Lambda, SQS, IAM permissions.
2. Java 17 & Maven
   • For running Spring Boot or other Java-based code.
3. Python 3.8+
   • Required for the face-recognition scripts, ffmpeg steps, etc.
4. Docker
   • For building container images for Lambda.

Setup & Installation

1. Clone this Repository

   git clone [https://github.com/<YourUsername>/aws-face-recognition.git](https://github.com/dvarshith/aws-face-recognition.git)
   cd aws-face-recognition
   

2. Configure AWS Credentials

   • Recommended: Use IAM Roles for EC2 and Lambda so you don’t store keys in code.
   • Or set environment variables locally (for testing):

     export AWS_ACCESS_KEY_ID=<YourAccessKey>
     export AWS_SECRET_ACCESS_KEY=<YourSecretKey>
     export AWS_DEFAULT_REGION=us-east-1
     

3. Build the Java Projects

   cd Project2
   mvn clean package
   

4. Deploy the Lambdas

   • For video-splitting and face-recognition (Project 3), either:
     • Upload the JARs to AWS Lambda via console, or
     • Build Docker images with their respective Dockerfile and push to ECR, then create Lambdas from those images.

Usage

1. Upload a Video (PaaS Example)
   • Upload test_00.mp4 to <ASU_ID>-input.
   • Lambda (video-splitting) extracts one frame, saves as test_00.jpg in <ASU_ID>-stage-1.
   • Lambda (face-recognition) runs face detection, saves the recognized name in test_00.txt in <ASU_ID>-output.
2. Web Tier (IaaS Example)
   • Run:

     java -jar target/Project2-0.0.1-SNAPSHOT.jar
     

   • Upload: POST / with form-data:

     inputFile=<image_or_video_file>
     

   • The API sends a message to SQS, triggers the app tier, and eventually returns the classification result.

Security Considerations

• Remove Hardcoded Credentials: Do not store AWS_ACCESS_KEY_ID or AWS_SECRET_ACCESS_KEY in Java/Python files.
• Use IAM Roles: Prefer instance profiles for EC2, and execution roles for Lambda.
• .gitignore: Make sure you aren’t committing .pem files, .log files with private data, or large training data sets.

Possible Improvements

• Add CI/CD (GitHub Actions or Jenkins) for automatic builds and deployments.
• More Granular IAM Policies to adhere to the principle of least privilege.
• Performance Tuning for your Lambda containers (memory, concurrency).
• CloudWatch Alarms & Metrics for deeper monitoring of queue length, CPU, memory usage, etc.

Acknowledgments

• Dataset, test cases, etc. provided by Vista Lab from Arizona State University [https://github.com/visa-lab/CSE546-Cloud-Computing/tree/main].

License

This project is released under the MIT License. That means you’re free to use, modify, and distribute the code, but you do so at your own risk.

Contact

Author: Varshith Dupati
GitHub: @dvarshith
Email: dvarshith942@gmail.com
Issues: Please open an issue on this repo if you have questions or find bugs.