AWS Kinesis real time video processing using sagemaker

Question

I am working on a POC for video face recognition using AWS kinesis, sagemaker. I have set up the environment to send my webcam video to kinesis. I am also working on a custom docker implementation of my face recognition code to be deployed in sagemaker.

I am stuck with few things here and need some help: I will be using python to build my applications.

1. How do i send kinesis video frames as input to sagemaker model as the input will be frame.
2. Once the above is done the custom sagemaker code will return the face detected as well as the bounding box. I can call this sagemaker end point from my custom chalice flask app to get the output but I am not sure how to get the frame back so that i can show it in my web app built using chalice-deployed in aws as well. The web app will stream back the video with bounding box.

I do not want to use amazon rekognition as well as HLS because they will incur a lot of cost.

Can anyone help me with the above problems or suggest a better approach to solve it. Language i can use is python.

Answer 1

Have you seen this reference implementation? https://docs.aws.amazon.com/kinesisvideostreams/latest/dg/examples-sagemaker.html