How can these requirements be met by using the LEAST amount of ongoing management overhead and causing MINIMAL disruption to the existing system?
Set up an AWS Storage Gateway, file gateway appliance on-premises. Use the MAM solution to extract the videos from the current archive and push them into the file gateway. Use the catalog of faces to build a collection in Amazon Rekognition. Build an AWS Lambda function that invokes the Rekognition Javascript SDK to have Rekognition pull the video from the Amazon S3 files backing the file gateway, retrieve the required metadata, and push the metadata into the MAM solution.
Set up an AWS Storage Gateway, tape gateway appliance on-premises. Use the MAM solution to extract the videos from the current archive and push them into the tape gateway. Use the catalog of faces to build a collection in Amazon Rekognition. Build an AWS Lambda function that invokes the Rekognition Javascript SDK to have Amazon Rekognition process the video in the tape gateway, retrieve the required metadata, and push the metadata into the MAM solution.
Configure a video ingestion stream by using Amazon Kinesis Video Streams. Use the catalog of faces to build a collection in Amazon Rekognition. Stream the videos from the MAM solution into Kinesis Video Streams. Configure Amazon Rekognition to process the streamed videos. Then, use a stream consumer to retrieve the required metadata, and push the metadata into the MAM solution. Configure the stream to store the videos in Amazon S3.
Set up an Amazon EC2 instance that runs the OpenCV libraries. Copy the videos, images, and face catalog from the on-premises library into an Amazon EBS volume mounted on this EC2 instance. Process the videos to retrieve the required metadata, and push the metadata into the MAM solution, while also copying the video files to an Amazon S3 bucket.
Explanations:
This option uses an AWS Storage Gateway (file gateway) to seamlessly transfer video files to Amazon S3 with minimal disruption. The integration with Amazon Rekognition allows for automated metadata extraction based on the video content, including faces, and the use of AWS Lambda minimizes ongoing management by automating the process of invoking Rekognition and updating the MAM system.
This option suggests using a tape gateway, which is not suitable for video file processing since it treats data as tape rather than files. Additionally, processing metadata from videos in a tape format would be inefficient and complex, leading to higher management overhead and potential disruptions to existing workflows.
While using Amazon Kinesis Video Streams for ingesting video content is a viable approach, it introduces unnecessary complexity and potential latency issues for a repository that already exists on tape. The streaming approach may not integrate smoothly with the current MAM system and requires additional management of Kinesis resources.
Setting up an EC2 instance with OpenCV requires significant management overhead and may lead to performance issues due to the need for manual processing. Copying videos and images to EBS adds complexity and does not utilize the native AWS capabilities for automated processing, which increases operational burdens and does not align with the goal of minimal disruption.