My team is developing a service that runs on AWS and ingests pictures from an S3 bucket. These pictures are frames from video streams from surveillance cameras, resulting in no Exif metadata for them, and most of the time, there would be no one in the pictures.
The service needs to 'read' these pictures and distinguish if any of them are upside down so that we can determine if those cameras are working properly. Initially, I thought we could use AWS Rekognition, but upon reading the documentation, it seems it cannot meet the requirement.
So, I'm wondering if there is any AWS service or library that could accomplish this and be deployed either in Lambda or ECS?