Preprocess input source

I have the following situation.
I don’t want to use direct video input streams, but preprocess them (e.g. with OpenCV functions) and then forward it to my deepstream pipeline.

Ideally, the image data is published on a broker and received by the pipeline.

Are there any ideas how to do that?

We have several types of sources supported in deepstream-app:

Please check if you can pick one for your usecase. If none can fit your usecase, you may need to modify the code of source bin to run

appsrc ! video/x-raw,format=NV12 ! nvvideoncovert ! 'video/x-raw(memory:NVMM),format=NV12'

I checked those option before… It seems that RTSP might be a way to solve it. So I assume, I have to write the pre-processed stream as RTSP. Any hints how to do that, if I want to pre-process the stream with OpenCV and then provide via RTSP to the deepstream pipeline.

You may leverage test-launch to run a rtsp source: