Gstreamer writing to CUDA memory and zero copy cv::cuda::GpuMat with Jetpack 5.1.2

[AastaLLL]

Hi,

Please check the two samples below:

Error generated while running the code after connecting the camera

Hi, Sorry that I missed one thing here. nvoverlaysink is deprecated on JetPack 5, and please use nv3dsink instead. launch_stream << "nvarguscamerasrc name=mysource ! " << "video/x-raw(memory:NVMM),width="<< w <<",height="<< h <<",framerate=30/1,format=NV12 ! " << "nvvidconv name=myconv ! " << "video/x-raw(memory:NVMM),format=RGBA ! " << "nv3dsink "; gst_cv_gpumat.tbz2 (3.4 KB)

OpenCV CUDA processing from gstreamer pipeline [JP4, JP5]

Just sharing here some findings from my experiments and some example code about how to use OpenCV cuda processing in gstreamer pipeline leveraging Jetson HW and SW. This is far from clean code… I’m just a stupid dog, be careful ;-) No warranty at all and no expected support ! This example builds a gstreamer app that will launch the pipeline string passed as argument. In that pipeline, the app will try to find an element with name=conv. Then it will add a src pad probe to it that will be ab…

It is expected to work on JetPack 6 as both use the NvBufSurface API.
Thanks.