• Hardware Platform (Jetson / GPU) Orin Devkit
• DeepStream Version 6.1
• JetPack Version (valid for Jetson only) 5.1
• TensorRT Version 8.4
So I am trying to record X seconds before and after a certain event (let’s say an obstacle is detected). How will I go about this and what are the elements I should use to achieve this?
Currently, to achieve this I am using a list to store frames (as numpy arrays). The number of frames I store is FPSX frames and continually update the list in FIFO fashion. Once an event is received I wait for the required frames (FPSX frames after the event) and then write them using OpenCV’s video writer by spawning a thread. It works well. Now I am looking to find a deep-stream way to achieve the same because maintaining a list of frames for tens of streams is not really practical and will lead to performance issues.
Also if it matters I am already recording raw streams and also publishing them as RTSP.
I have attached the pipeline for a better understanding of the workflow that I currently have.
Edit: So I am thinking ‘valve’ with a ‘splitmuxsink’ would be the way to go. Any pointers?
pipeline.pdf (49.4 KB)