Building an automated serverless facial analysis system.

2024-05-16 3 min read Projects Barny Baron

Journey: 📊 Community Builder 📊

Subject matter: Building on AWS

Task: Building an automated serverless facial analysis system.

This project is about Performance, Availability, and receiving real-time notifications when new analysis is performed.

Using the 6 Pillars of the AWS Well-Architected Framework, Performance Efficiency and Sustainability will be achieved in this build.

I will be setting up a Serverless Machine Learning service to analyse new image uploads and then perform real-time notifications to the Administrator of its findings.

Machine Learning

Credit: This serverless architecture was created using some guidance from Edithcherotich on Medium Here.

What did I use to build this environment?

Amazon S3
AWS Lambda
Amazon Rekognition
Amazon DynamoDB
Amazon SNS
AWS IAM
Amazon CloudWatch

What is built?

A new S3 bucket which triggers a Lambda Function
Amazon Rekognition is used to detect faces in newly uploaded images
An SNS Topic to receive calls for email distribution
An SNS Subscription to the new Topic
A DynamoDB Table for Amazon Rekognition to populate data into if a face is detected
An SNS trigger to notify an Admin if a face is detected, confirming the expected gender and age

This task involved using Amazon Rekognition to analyse new images that are uploaded to an S3 bucket. AWS Lambda triggers the analysis. If a face is detected by Amazon Rekognition, it determines the gender and age of the person before then populating it into a DynamoDB table. Lambda then triggers an SNS notification via email to the Account Administrator.

I also used the following AWS documentation:

https://docs.aws.amazon.com/sns/latest/dg/sns-getting-started.html

https://docs.aws.amazon.com/sns/latest/dg/sns-create-topic.html

https://docs.aws.amazon.com/sns/latest/dg/sns-create-subscribe-endpoint-to-topic.html

Once I had configured all of the areas, I uploaded a photo of a person into my S3 bucket with the titles ‘celebrating.jpg’ and ‘professional.jpg’ so as not to show gender bias. AWS Lambda triggered off the upload and Amazon Rekognition confirmed a face was present, identified the likely gender and age, and sent a real-time notification to my mailbox.

As a further test, I then uploaded a photo of ‘coffee.jpg’ to my S3 bucket and did not receive any email.

When checking the CloudWatch logs, I identified an error that ‘gender could not be defined’, which is expected behaviour. If we wanted to take the code further, error handling could be introduced to send an email notification of a non-human image detected.