I want to run some machine learning models (like object detection) on a live video stream from multiple cameras ( thousands of cameras) and show all cameras' playback in a web application where the user can show all cameras' feeds at a time or if the user wants can show only one camera's feed. I plan to deploy this application in AWS.
My plan is to use the Amazon Kinesis data stream for managing multiple cameras and then run an EC2 instance to read the data stream from kinesis and run all machine learning processes on every stream. But I do not find any way to show all cameras' feeds to a web application with drawing bounding boxes on every frame of the feed.
What is the working process or architecture of this scenario so that I can collect multiple camera streams and run the Machine Learning process on every stream and finally show all camera feed to a web application?