- Newest
- Most votes
- Most comments
Hello Ayan,
First off, kudos for developing your own video enhancement solution! Now, if you are not going to utilize the features of MediaLive to process and distribute the video, you might just run all of your processes within EC2. I am going to post in an example from Github, which has some of the features I am recommending. The workflow is for a live subtitling workflow, so it isn't completely on point - and it is deprecated due to some underlying elements written in an old version of Python. However, I think if you backward engineer the process, you can modify the workflow to run your enhancement software on the video and stream it live. Here is what the diagram looks like as published:
You could use OBS to contribute video to Ngenex, which would work as your ingest server. In the above workflow, this is how the custom application is run to modify the video and insert subtitles: "When a video publish begins, the Nginx RTMP module will execute a shell script that has been pre-installed on the EC2 instance. This script kicks off a video production workflow to inject subtitles, using ffmpeg, NodeJS, and a modified version of libcaption that is hosted in this repo." (verbiage from the Github README)
You could use a workflow like this to have your video enhancement algorithm run in NodeJS, then send the enhanced live video via CloudFront, or whatever CDN you choose. You could also modify the workflow to save a copy of the source video to S3, so you could retain the original mezzanine level video for future processing and use in VOD workflows. You could also keep the subtitle insertion functionality, if you wish.
Also, how much latency can you tolerate in your workflow? Is it a simple broadcast with no interactive timed metadata? If so, no problem. But, running this workflow on live video streams will introduce some level of latency. You will have to test and see what that latency will amount to, in real-life.
I hope this brainstorming session helps you with your project. Please feel free to tell us more of your parameters, and maybe we can tighten up our suggestions. I'll keep looking for more examples of relevant workflows and I will update this thread when I find more content.
Good luck in your building efforts!
Hi,
Before going into proprietary developments, I'd suggest you to study AWS Elemental MediaConvert. See https://aws.amazon.com/mediaconvert/
It seems to do a lot of the things that you want to achieve in terms of automated video processing and conversion.
Best,
Didier
I have just gone through the media convert documentation on a high level. As far as I understood, it's for transcoding video. I didn't got any reference on how to connect EC2 with this service to trigger my video enhancement algorithm on the live stream. Any reference on that will be helpful.
Thanks, Ayan
MediaConvert can ingest live HLS sources but outputs only VODs. Jobs are time-clipped to the most recent segment advertised in the source manifest at the start of the job. Jobs run asynchronously. Not a live stream processing solution.
The simplest answer is to send either a packet stream (rtmp or rtp-fec) or upload an HLS file group from MediaLive to the EC2 instance. A listener process of some kind will need to run on the instance, such as ffmpeg or your own code if ingesting a bitstream; or an HTTP server for receiving a file group. For live content processing you shouldn't need to transiently store more than a few minutes of content at time. The allocated disk should be fast but need not be large.
Relevant content
- Accepted Answerasked a year ago
- asked 4 days ago
AWS OFFICIALUpdated a year ago- AWS OFFICIALUpdated 5 months ago
- AWS OFFICIALUpdated a year ago
Hey thanks. This approach really seems feasible. Thanks for this detailed explanation and link to the repo. I am looking for latency less than 5-6 secs at least for the first iteration as the algorithm intend to be use for real time security surveillance purpose. I will try the above mention approach and update the outcome here. Thanks!!
Hi @BCole2019, thanks for your suggestion. I have tried to follow a similar approach as mentioned above for my use case. Finally, I am able to process the live stream in an EC2 instance and then distribute!! I have setup up a NGINX RTMP server for ingestion which trigger a script to process the video frames and finally send to AWS IVS for playback. Though currently I am seeing a significant latency of upto 1 min from source to playback, it may be due to ffmpeg configuration which I believe, I need to tune for my use case. I would also like to hear any suggestion from you to improve the latency.