Remove logos or static overlays from a live stream
Utilizing CUDA hardware decoder cuvid each frame is decoded and copied to a cuda::GpuMat and saved in frames queue, the frames queue is processed when it reaches the set value of pfrm (200 by default), once the queue is filled it's passed to process_video_frame and using cv::addWeighted all the queued frames are added with alpha value 0.1, this way the created a single frame the shows clearly the desired template that should be matched in process_vframe, because matching the template frame by frame is much slower and less accurate, once a template detected cv::inpaint applied to add smooth blurring over the object and essentially hiding it in the output stream
Ubuntu
sudo apt install libopencv-dev libavcodec-dev libavdevice-dev libavfilter-dev libavformat-dev libavutil-dev
OR
git clone https://github.com/Marware/opencv-ffmpeg-cuda-build
Follow the insructions in the repo
Using opencv-ffmpeg-cuda-build will get and build the latest code from OpenCV and FFmpeg codebases
cd mdelogo
qmake
make
./mdelogo input=[URL/FILE] output=[URL/FILE]
- It the feature to detect text using OCR
Tesseractis available but commented due to limited testing - The ability to use CPU-only processing in decoding, encoding and detection is available, needs more optimization
- Contributions are welcome!