Motion Sensing with Raspberry Pi Camera and Cat Face Scanning with AWS Lambda + OpenCV

I wrote a Python script that detects camera motion with a Raspberry Pi camera and uploads photos to S3. The frames are then searched for cat faces using OpenCV running on AWS to alert them of heightened 'cattivity' in the house.

This could also be done using Amazon’s cool new image ‘Rekognition’ service, but it would cost more if you process thousands of frames daily.

I wanted to know how often my flatmate's cat attempted to attack his mouse enclosure. Cat detection events are sent to DataDog using the StatsD protocol for alerting!

Raspberry Pi Agent Code

I am monitoring the occurrence of motion events using DataDog and uploading the frames to S3 to detect further objects in the frame, such as cats or people.

The full source code is on GitHub: timatooth/catscanface

Haar-Cascade Cat Detection with AWS Lambda

You will need to package compiled Numpy & OpenCV libraries prepared for running inside the AWS Lambda functions thanks to aeddi https://github.com/aeddi/aws-lambda-python-opencv

Update: 2019 I Believen numpy & OpenCV may work out-of-box using AWS SAM to package Lambda functions.

from __future__ import print_function
import logging
logger = logging.getLogger()
logger.setLevel(logging.INFO)
import os
import os.path
import numpy
import cv2
import boto3

# Config
s3_prefix = 'catscanface/'
s3_bucket = 'timatooth'

cat_cascade = cv2.CascadeClassifier('haarcascade_frontalcatface_extended.xml')
s3_client = boto3.client('s3')

def scan_frame(image_data):
    img_array = numpy.asarray(bytearray(image_data), dtype=numpy.uint8)
    frame = cv2.imdecode(img_array, 0)
    cat_faces = cat_cascade.detectMultiScale(frame, 1.3, 5)
    # Draw rectangle onto region where cat face is found
    cat_detected = False
    for (x, y, w, h) in cat_faces:
        cat_detected = True
        cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
        logger.info("Found cat")

    if cat_detected:
        _, buf = cv2.imencode(".jpg", frame)
        return buf.tobytes()


def get_image(object_key):
    response = s3_client.get_object(Bucket=s3_bucket, Key=object_key)
    image_bytes = response['Body'].read()
    logger.info('Got {} bytes'.format(len(image_bytes)))
    return image_bytes


def lambda_handler(event, context):
    object_key = event['Records'][0]['s3']['object']['key']
    image_bytes = get_image(object_key)
    cat_image = scan_frame(image_bytes)
    if cat_image is not None:
        key = s3_prefix + 'detections/' + os.path.basename(object_key)
        logger.info('Saving cat detection image: {}'.format(key))
        s3_client.put_object(
            Bucket=s3_bucket,
            Key=key,
            Body=cat_image
        )

CloudFormation

  ---
AWSTemplateFormatVersion: '2010-09-09'
Description: Cat facial recognition OpenCV lambda
Parameters:
  S3BucketName:
    Type: String
    Description: The S3 bucket which contains cat pictures
  S3BucketPrefix:
    Type: String
    Description: A bucket prefix where the cat pictures are found
    Default: /catscanface
Resources:
  LambdaRole:
    Type: AWS::IAM::Role
    Properties:
      AssumeRolePolicyDocument:
        Version: '2012-10-17'
        Statement:
        - Effect: Allow
          Principal:
            Service:
            - lambda.amazonaws.com
          Action:
          - sts:AssumeRole
      Path: "/"
      Policies:
      - PolicyName: CloudwatchLog
        PolicyDocument:
          Version: '2012-10-17'
          Statement:
          - Effect: Allow
            Action:
            - logs:CreateLogGroup
            - logs:CreateLogStream
            - logs:PutLogEvents
            Resource: arn:aws:logs:*:*:*
      - PolicyName: AccessCatPictureBucket
        PolicyDocument:
          Version: '2012-10-17'
          Statement:
          - Effect: Allow
            Action:
            - s3:GetObject
            Resource: !Join ['', ['arn:aws:s3:::', Ref: S3BucketName, Ref: S3BucketPrefix, '/motion/*']]
      - PolicyName: WriteDetectedCatFrames
        PolicyDocument:
          Version: '2012-10-17'
          Statement:
          - Effect: Allow
            Action:
            - s3:PutObject
            Resource: !Join ['', ['arn:aws:s3:::', Ref: S3BucketName, Ref: S3BucketPrefix, '/detections/*']]
  Lambda:
    Type: AWS::Lambda::Function
    Properties:
      Code:
        ZipFile: 'print("hello world")'
      Description: Scan images uploaded to S3 for cats
      FunctionName:
        Ref: AWS::StackName
      Handler: catscan.lambda_handler
      MemorySize: 256
      Role: !GetAtt LambdaRole.Arn
      Runtime: python2.7
      Timeout: '4'
Outputs:
  LambdaRole:
    Description: IAM Role for LambdaRole
    Value:
      Ref: LambdaRole
  Lambda:
    Value: !GetAtt Lambda.Arn