I’m wondering how to construct a deepstream pipeline where a file is written based on an object detection.
For example a scenario like this:
When a person is detected, start writing the stream to a file for 30 seconds then stop. When a person detection event occurs again start writing another file.
Such that you end up with a whole series of 30 second video clips showing all the person detection’s.
There are allot of potential use-cases for deepstream and edge devices where the video stream is only sent to the cloud on a positive detection/event but I can’t see how to do this as all the samples are designed around constant streaming right through the pipeline.
Another potential example would be with a jetson nano setup with a camera and watching for birds. When a bird is detected it starts streaming the video to the cloud - instead of all the examples here which would just fulltime stream regardless of whether any detection exist and using up bandwidth.
Just after some ideas to continue my research? ;-)