about

video stream hardware decoding and color space conversion by ffmpeg and cuda.

brief

This project demonstrate how to decode video stream and convert color space by using ffmpeg and GPU. When hardware decoding video stream, the project use ffmpeg(version: 3.3) and h264_cuvid codec(based on NVIDA GPU). When converting NV12 to BGR24 color space, the project use CUDA(version: 8.0) api where readers can find source conversion codes.

As this is a part of Video Monitor project focus on decoding RTSP video stream to OpenCV cv::gpu::GpuMat frame, high performance is a must. By using multiple GPUs, we can reduce stress of CPUs and most important, we can increase the processing rate.

build

make

run

./server

dependency

• ffmpeg 3.3
• cuda 8.0
• opencv 2.4.13

export CMAKE_PREFIX_PATH=/path/to/opencv/dir
export LD_LIBRARY_PATH=$CMAKE_PREFIX_PATH/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/path/to/ffmpeg/lib:$LD_LIBRARY_PATH