# Amazon Rekognition People Label Detection via KVS
This is the README for the whole people detection pipeline designed for real-time video analysis.

There are three Lambda functions placed inside this repository:
### 1. Kinesis Video Stream to Rekognition (`cctv-people-rekogniton`)
This function utilizes Amazon Rekognition's stream processor, which is configured to monitor an incoming Amazon Kinesis Video Stream for people label detection.

Note that this Lambda function is intended to process **a single Kinesis video stream only**. For a multi-camera setup, multiple deployments of this Lambda function are necessary.

This is built on top of AWS KVS Consumer Library for Python. The original repository can be found here: https://github.com/aws-samples/amazon-kinesis-video-streams-consumer-library-for-python/

### 2. SNS to DynamoDB and OpenSearch (`cctv-people-sns-dynamodb`)
This function writes a new entry to DynamoDB and Opensearch tables whenever there is a detection coming from Rekognition via SNS.

### 3. DynamoDB to Supabase (`cctv-people-dynamodb-supabase`)
This function acts an extension of the previous function, writing a new entry to Supabase whenever there is a new entry in DynamoDB.

## Prerequites
The following must already be set up before deploying this pipeline:
1. A working Kinesis video stream as input
2. An S3 Bucket where all frames with detection go
3. A SNS topic that will receive the notification coming from Rekognition
4. A DynamoDB table
5. An OpenSearch Service collection

It is advised to have all above components in the same region for easy configuration.

## Setup
### Part I. Kinesis Video Stream to Rekognition (`\cctv-people-rekogniton`)

1. Create an IAM role for the Rekognition stream processor. This role must contain the following AWS-managed policies:
    - AmazonRekognitionServiceRole
    - AmazonS3FullAccess

2. Provide the following details in `lambda_function.py`:
    - Kinesis video stream name and ARN
    - Stream processor name (must be unique to the KVS stream)
    - S3 bucket name
    - IAM role ARN
    - SNS topic ARN

3. Package the folder as a ZIP file.
4. Create an AWS Lambda function and upload the ZIP file.
5. Add an EventBridge trigger with the schedule expression of `rate(5 minutes)`

### Part II. SNS to DynamoDB and OpenSearch (`\cctv-people-sns-dynamodb`)

1. Create a Lambda function and upload the `lambda_function.py` file.
2. Add a trigger to receive notifications from the SNS topic.
3. Add the following AWS-managed permissions to this Lambda function:
    - AmazonDynamoDBFullAccess_v2
    - AmazonOpenSearchIngestionFullAccess
    - AmazonOpenSearchServiceFullAccess
    - AmazonS3FullAccess
4. Add an inline policy for OpenSearch Serverless:
```
{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "Statement1",
            "Effect": "Allow",
            "Action": "aoss:*",
            "Resource": "*"
        }
    ]
}
```
5. Create an AWS Managed Grafana workspace.
![alt text](./imgs/grafana.png)
![alt text](./imgs/grafana-interface.png)

6. Access the workspace and install opensearch aws sig4 and opensearch extension.

![alt text](./imgs/datasource1.png)
![alt text](./imgs/datasoruce2.png)

7. Point it to the opensearch url and make sure Aws SigV4 is enabled.
Configure the correct index.

### 3. DynamoDB to Supabase (`\cctv-people-dynamodb-supabase`)
1. On a local machine, prepare a folder for Supabase installation.
2. Open a terminal pointed at this directory and run the following command:
```
pip3 install supabase --platform manylinux2014_x86_64 --python-version 3.12 --only-binary=:all: -t . 
```
3. Place the `lambda_function.py` inside this folder.
4. Package the folder as a ZIP file.
5. Create an AWS Lambda function and upload the ZIP file.
6. Add the following AWS-managed permission to the IAM role for this Lambda:
    - AmazonDynamoDBFullAccess_v2
7. Enable stream on the DynamoDB table. Set view type to **New image**.
8. Add trigger to the Lambda function to connect to the DynamoDB stream.