Implementation of a Pipeline for Analyzing the Interview videos using Amazon Rekognition Service

Industry

Software and Internet

Expertise

Amazon S3, AWS Fargate, Amazon Athena, Amazon DynamoDB, AWS Lambda

Offerings/solutions

Optimized Video Analysis with AWS Services for Efficient Frame Processing and Storage

About the Client

The client is an Indian political party, a non-profit organization with over 10,000 employees and 484 associated members. It has a strong historical legacy, focusing on national development, and has successfully governed the nation multiple times, with a robust presence in state legislatures.

Highlights

30%

Efficiency Improvement

40%

Storage Cost Reduction

25%

False Positive Reduction by splitting frames

The Challenge

The objective is to provide real time insights by analyzing the interview recorded videos using AWS Machine Learning Service. The API response gives the details of the number of users, face and eye direction of those users, added flags if more than 2 users are detected, detects objects and unsafe contents from that video, and stores that response in Amazon S3.

I appreciate the team efforts in showing the real-time insights on this project, support, professionalism, timeliness, and results shown by the CloudThat team.

Navin GV, IDesk Technologies Pvt. Ltd.

Solutions

Created an Amazon EC2 instance for running the docker image and pushing that docker image to the Amazon ECR repository.
Implemented Docker to run the script for slicing the videos into the frames and each frame gets stored in the bucket.
Used AWS Lambda service for running the Amazon Rekognition APIs for each frame, and the SQS triggers this lambda by passing the frame details to it when the frame uploads into the bucket.
By doing the above step, it would give the preprocessed data by analysing all the API responses, and those response gets stored in the bucket.
Created a crawler in the AWS Glue for transferring the responses of all the frames and the crawler runs when the responses uploading in the bucket is completed.
Implemented SQL query to analyze all the responses of all frames of a video for identifying the face angle detection, eye gaze detection, object detection, and new person detected in the video.
Used a new AWS Lambda for storing the Amazon Athena aggregated resultsinto the dynamodb table.
We built a pipeline for analyzing the videos and storing the Amazon Athena results in the Dynamodb table. The pipeline starts when the video comes into the bucket.
AWS Services used are Amazon ECS, Amazon VPC, Amazon ECR, Amazon S3, Amazon DynamoDB, Amazon SQS, Amazon CloudWatch, AWS IAM, AWS Lambda, and Amazon Rekognition.