This project implements a facial recognition pipeline for a building's security system. It analyzes images captured by a front-door camera, processing up to 500 employee scans daily. Using Amazon Rekognition, the system detects faces in images uploaded to S3. If a face is recognized, the results are stored in DynamoDB, and a notification is sent via SNS to authorize security to unlock the door, allowing employees to enter the building and begin work.
source:https://tutorialsdojo.com/5-best-cloud-projects-for-beginners/
- Amazon S3
- AWS Lambda
- Amazon Rekognition
- Amazon DynamoDB
- Amazon SNS
US East (Ohio) - us-east-2
-
A user uploads an image from their computer into an Amazon S3 bucket.
-
The bucket is configured with event notifications so that whenever a new image is uploaded, it automatically triggers the AWS Lambda function.
S3 Standard - General purpose storage for any type of data, typically used for frequently accessed data First 50 TB / Month $0.023 per GB
- PUT, COPY, POST, LIST requests $0.005/per 1,000 requests
- GET, SELECT, and all other request $0.0004/per 1,000 requests
- 1️⃣ S3 Storage
15,000 images × 0.5 MB = 7,500 MB = 7.5 GB stored per month. Pricing (S3 Standard): $0.023 per GB Cost: 7.5 × $0.023 = $0.1725 ≈ $0.18 per month
- 2️⃣ S3 Requests
PUT (uploads) = 15,000 / 1,000 × $0.005 = $0.075 GET (retrieval by Rekognition) = 15,000 / 1,000 × $0.0004 = $0.006 Total = $0.081 ≈ $0.08 per month
- Lambda function is invoked when the new object (image) is created in S3.
import boto3
import json
from decimal import Decimal
rekognition = boto3.client('rekognition', region_name='us-east-2')
dynamodb = boto3.resource('dynamodb', region_name='us-east-2')
sns = boto3.client('sns', region_name='us-east-2')
DYNAMO_TABLE = 'FaceMetadata'
SNS_TOPIC_ARN = 'arn:aws:sns:us-east-2:094092120892:FaceDetectedTopic'
REKOGNITION_COLLECTION = 'employeeFaces'
def lambda_handler(event, context):
try:
record = event['Records'][0]
bucket = record['s3']['bucket']['name']
key = record['s3']['object']['key']
print(f"Processing image: {key} from bucket: {bucket}")
# Step 1: Detect faces
response = rekognition.detect_faces(
Image={'S3Object': {'Bucket': bucket, 'Name': key}},
Attributes=['ALL']
)
face_details = response.get('FaceDetails', [])
if not face_details:
print("No face detected.")
return {'statusCode': 200, 'body': 'No face detected.'}
face = face_details[0]
print("Face detected. Writing to DynamoDB...")
# Step 2: Write to DynamoDB
table = dynamodb.Table(DYNAMO_TABLE)
table.put_item(Item={
'FaceId': key,
'AgeRange': {
'Low': Decimal(str(face['AgeRange']['Low'])),
'High': Decimal(str(face['AgeRange']['High']))
},
'Gender': {
'Value': face['Gender']['Value'],
'Confidence': Decimal(str(face['Gender']['Confidence']))
},
'Emotions': [
{
'Type': e['Type'],
'Confidence': Decimal(str(e['Confidence']))
} for e in face['Emotions']
]
})
# Step 3: Search for face match
match_response = rekognition.search_faces_by_image(
CollectionId=REKOGNITION_COLLECTION,
Image={'S3Object': {'Bucket': bucket, 'Name': key}},
MaxFaces=1,
FaceMatchThreshold=90
)
matches = match_response.get('FaceMatches', [])
match_info = matches[0]['Face']['ExternalImageId'] if matches else 'No match found'
print(f"Match result: {match_info}")
# Step 4: Publish to SNS
sns.publish(
TopicArn=SNS_TOPIC_ARN,
Message=f"Face detected in image: {key}\nMatch result: {match_info}",
Subject="Face Recognition Alert"
)
print("SNS notification sent.")
return {'statusCode': 200, 'body': f'Face processed. Match result: {match_info}'}
except Exception as e:
print(f"Error: {str(e)}")
return {'statusCode': 500, 'body': f'Error: {str(e)}'}The Lambda free tier includes 1M free requests per month and 400,000 GB-seconds of compute time per month.
-
Inside the Lambda function, you call Amazon Rekognition.
-
Rekognition analyzes the uploaded image and checks whether it contains faces.
Group 1 AssociateFaces First 1 million images $0.001
- workload: 500/day × 30 days = 15,000 images/month.
- 15,000 × $0.001 = $15.00/month (ignoring any free-tier credits).
The system sends SNS notifications for ALL scenarios:
- Notification: Sent immediately
- Subject: "🚫 Face Recognition Alert - No Face Detected"
- Action: Access denied, door remains locked
- Notification: Sent with employee details
- Subject: "✅ Face Recognition - Authorized Access"
- Action: Door unlocked, employee authorized to enter
- Data Stored: Employee ID, match confidence, face metadata
- Notification: Sent immediately (
⚠️ SECURITY ALERT) - Subject: "🚫 Face Recognition - Unauthorized Access Attempt"
- Action: Door remains locked, security notified
- Data Stored: Face metadata for investigation
- Notification: Sent if any error occurs
- Subject: "❌ Face Recognition - Processing Error"
- Action: Requires manual investigation
- The Lambda function writes metadata to DynamoDB for all processed images:
- Match status (MATCHED/UNMATCHED)
- Employee ID (if matched)
- Match confidence score
- Face attributes (age, gender, emotions)
- Processing timestamp
DynamoDB Standard table class > On-Demand Throughput Type DynamoDB Monthly Cost Estimate 500 uploads/day × 30 days = 15,000 writes/month
- Writes = ~$0.02
- Reads = ~$0.003 (depends on usage)
- Storage = ~$0.004
- Total ≈ $0.03/month
✅ AUTHORIZED ACCESS - Employee Recognized
Image: employee_photo.jpg
Bucket: rekognition-upload-bucket1
Employee ID: EMP001
Match Confidence: 95.50%
Status: MATCHED
Action: Door should be UNLOCKED. Employee is authorized to enter.
🚫 UNAUTHORIZED ACCESS - Unknown Person
Image: unknown_person.jpg
Bucket: rekognition-upload-bucket1
Status: UNMATCHED
Match Result: No matching employee found in database
Action: Door should remain LOCKED. Unauthorized access attempt detected.
Security should be notified immediately.
Image Processing Result
Image: invalid_image.jpg
Status: No face detected
The uploaded image does not contain any detectable faces.
Access should be denied.
$2 per 100,000 emails → your 15,000/month is only $0.30/month.
Tracks metrics for:
- Lambda: invocations, duration, errors
- S3: object count, storage size
- DynamoDB: read/write capacity, throttling
- SNS: messages published/delivered/failed