Image_Recognition-as-a-Service

Leveraging AWS Lambda to create a Video Analysis Application

Summary

In this project, I developed an elastic cloud application using AWS Lambda and other PaaS cloud services. The goal was to create an application capable of automatic scaling and cost-effectiveness through serverless computing. AWS Lambda, a leading function-based serverless computing service, was utilized to build a more sophisticated cloud application compared to Project 1. This project aimed to provide a meaningful cloud service to users, while the skills and techniques learned are foundational for future cloud development endeavors.

Description

The cloud application is a video analysis platform employing Lambda function and ECR to execute a multi-stage processing pipeline for user-uploaded videos:

1. Pipeline Start (Stage 1):

   • Users upload videos to the designated input bucket, named input.
   • Upon upload, the video-splitting Lambda function is triggered.

2. Stage 1 - Video Splitting:

   • The video-splitting function utilizes FFmpeg to split videos into frames.
   • It saves these frames as JPG images in the stage-1 bucket.
   • This function triggers the face-recognition function asynchronously upon completing its task.

3. Stage 2 - Face Recognition:

   • The face-recognition function processes the frames from Stage 1.
   • Using a Single Shot MultiBox Detector (SSD) algorithm and a pre-trained CNN model (ResNet-34), it identifies faces in the frames.
   • Recognized faces are stored as text files in the output bucket, with each file containing the name of the identified person.

Architecture Diagram

Bucket Details

• Input Bucket (input):

  • Stores uploaded .mp4 video files.
  • Triggers the video-splitting Lambda function upon new uploads.

• Stage-1 Bucket (stage-1):

  • Stores JPG images resulting from the video-splitting function.
  • Names of images correspond exactly to the input video names with .jpg extension.

• Output Bucket (output):

  • Stores text files (.txt) containing recognized names of faces from Stage 2.
  • Names of text files match the original input video names with .txt extension.

Function Details

• Video-splitting Function:

  • Triggered by new video uploads to input.
  • Uses FFmpeg to split videos into frames, storing them in stage-1.
  • Asynchronously invokes the face-recognition function upon completion.

• Face-recognition Function:

  • Triggered by the completion of video-splitting.
  • Processes each image frame to detect and recognize faces using OpenCV and ResNet-34.
  • Stores recognized names as text files in output.

AWS Services used:

1. IAM
2. S3
3. Lambda
4. Cloud Watch
5. ECR

---

This project not only enhanced my proficiency in cloud programming and PaaS but also equipped me with practical skills in implementing scalable and efficient serverless applications using AWS Lambda.