AWS Kinesis Video stream use case

I wanted to check one thing, we are storing live video in the Kinesis video stream and the next step is to push/pull the data to our own object detection component. I found two way of achieving this.

1. Using HLS based using streaming session URL from this we can use video stream data into our own custom object detection module.

2. Other one is write consumer using AWS parser lib we can fetch the video stream frame by frame and use it in the object detection.

So wanted to know which is best suitable for real time video streaming data.

Thanks, Venkat