What is the best way to decode a MJPEG stream by using GPU to feed an OpenCV application?

Gaming and Visualization Technologies Video Processing & Optical Flow
1

Hi everyone,
I have wrote an OpenCV application in Qt (MSVC 2017) to decode raw frames from a MJPEG steam for further process, now I am trying to use GPU for decoding the stream. what is the best and easiest way to do so?

2

Hi,

You can use NVIDIA Video Codec SDK. Please download the package from NVIDIA VIDEO CODEC SDK | NVIDIA Developer. We have samples which illustrate JPEG decoding, which you can refer to.

Let us know if you need more help.

Thanks,
Ryan Park

3

Hi,

Thank you, I used the mentioned SDK and recompiled OpenCV with NVCUVID support. now I can use opencv gpu video decoder :).

4

I am trying to decode a MJPEG frame from a camera.

@pouya.ahmadvand - can you please elaborate recompiling opencv with NVCUVID ?

what are the switches/options you used? my understanding was NVCUVID was deprecated.

also, for the cuda gpu decoder you are referring to, are you using cudacodec::VideoReader ?

finally are you reading video files or reading from camera?

thanks a lot for any help.

5

Yes, you must download nvidia codec sdk separately and recompile opencv (4.0.1) by NVCUVID. You can follow this link :

https://github.com/opencv/opencv_contrib/pull/1946

Yes I’m using cudacodec::VideoReader to decode a camera stream

6

Thanks!

did you implement your own cudacodec::RawVideoSource or how are you reading from camera?

is it USB or RTSP stream or other camera sources?