Jetson nano: Deepstream4 I need 4 camera streams,any kind will do

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I am working on a project that needs 4 camera streams. CSI camera only has 1 input. The bandwidth of the USB ports only allows for max 2 webcams to be processed. Anybody got any suggestions?

2

Use rtsp/simiar cameras and use the rtspsrc
or similar to get your video into the gstreamer pipeline. Nvidia’s Accelerated Gstreamer guide has some examples, iirc, but I could be mistaken. In any case, you will need that guide to know what elements to put in the pipeline and how. Many common GStreamer elements have accelerated versions you will want to use for performance sake. The DeepStream 4 package has some working rtsp examples as well. It you installed it, the examples are under /opt/nvidia/deepstream

3

Thanks for the Help
I will remember you at Christmas

4

Yw. I forgot to mention this guide here as well. It lists all the DeepStream components and their parameters. Of interest to you if you design your own pipeline are:

  • nvstreammux
    • to combine multiple streams into batches and attach batch metadata upstream required for:
  • nvinfer
    • which actually does the inference, attaches frame and object level metadata to the batch metadata, and passes them downstream to something like:
  • nvosd
    • to display your video and draw things on top of it like bounding boxes of detected objects.

    … and lots more useful tools, like to demux and export/stream the metadata.

    All (i think) of these are used in the DeepStream examples in /opt/nvidia/deepstream