Real-time Face Recognition using AWS Lambda, Docker, and Raspberry Pi

This project aims to build a distributed cloud application that utilizes Platform as a Service (PaaS) components and Internet of Things (IoT) devices to perform real-time face recognition on videos captured by the devices. This is partly an application of Edge Computing, which is a form of computing that is done near where the data is sourced, minimizing the need to be processed in a remote data center. This project specifically uses Raspberry Pi based IoT, AWS services such as Lambda, the first Function as a Service (FaaS) platform which facilitates serverless computing, as well as S3 and DynamoDB for storage needs, and Docker Engine for creating containers.

Setup and Execution

S3:

• Create an S3 bucket (we used ‘cc-project-videos’ as the bucket name’) to upload every 0.5-second video from Raspberry Pi.

DynamoDB:

• Create a DynamoDB table with ‘label’ as the partition key (we created a table named ‘student_table’).

• The items under the ‘label’ attribute should match the labels obtained from the machine learning model.

• Add other student information such as name, major, and year to the table.

Docker image creation:

• Build the docker image using the command given below specifying the AWS parameters which will be used to access AWS services like S3 and DynamoDB.

  docker build -t my_container --build-arg AWS_DEFAULT_REGION="us-east-1" --build-arg AWS_ACCESS_KEY_ID="your_id" --build-arg AWS_SECRET_ACCESS_KEY="your_key" .

• To test the docker image locally without deploying to lambda:

  • Run the docker image locally in a command prompt using the command:

    docker run --rm -p 9000:8080 my_container

  • Then, invoke the lambda function locally by using http post requests targeted to the URL specified below, passing a test image as the body. A sample test method is given in http_post_local_testing.py in the project files.

    http://localhost:9000/2015-03-31/functions/function/invocations

Lambda:

• Upload the created Docker image to Amazon ECR.

• Create a Lambda function in AWS console using the docker image from the ECR repository.

• Increase the Memory Configuration of Lambda for faster processing.

API Gateway:

• Create an API gateway(REST) and use that as a trigger for the lambda function.

• Enable Lambda Proxy integration.

• Include 'image/jpeg' as Binary Media Type under the Settings tab.

Raspberry Pi:

• Flash the Raspbian OS onto an SD card using a Raspberry Pi Imager and install it to the Raspberry device.

• Connect Raspberry Pi to a monitor using HDMI and use peripheral devices to boot up the device and connect it to a WiFi network (or use serial connection to do the same).

• Set up an SSH server on the device and use a laptop to connect to the device from thereon.

• Copy the capture_video.py file to Raspberry Pi and execute the file by giving arguments for the video time interval and total duration in seconds (e.g., python capture_video.py 0.5 300)