 |
 |
|---|---|
 |
 |
This project uses a pretrained Mask R-CNN from Detectron2 to compute the segmentation mask of a person in the shot and follows it using the OpenCV MOSSE tracker. The person bounding box and segmentation mask are updated accordingly with the Mask R-CNN throughput.
To install the project you need to clone it
git clone https://github.com/galatolofederico/rcnn-opencv-live-webcam-background-removal.git && cd rcnn-opencv-live-webcam-background-removal
And install the dependencies in a conda virtualenv
conda env create -f environment.yml
To run the project firstly you need to activate the conda virtualenv
conda activate webcam-background-removal
And then you can see the output with
python run.py --window
You can output the raw video to the stdout with
python run.py --raw
Outputting the video to VLC
python run.py --raw | vlc --demux=rawvideo --rawvid-fps=30 --rawvid-width=640 --rawvid-height=480 --rawvid-chroma=RV24 - --sout '#display'
Streaming the video via UDP
python run.py --raw --bgr | ffmpeg -f rawvideo -pix_fmt rgb24 -s 640x480 -r 30 -i - -an -f mpegts udp://0.0.0.0:5555
Forwarding the video to a virtual webcam with v4l2loopback
python run.py --raw --bgr | ffmpeg -f rawvideo -pix_fmt bgr24 -s 640x480 -r 30 -i - -pix_fmt yuv420p -threads 0 -f v4l2 /dev/video2
If you have a gcc version higher than 8 you need to set CC and CCX before installing Detectron2
CC=gcc-8 CCX=g++-8 pip install git+https://github.com/facebookresearch/detectron2.git
This code is released under the GPLv3, and it is free and open source as all the code should be. Feel free to do whatever you want with it :D.
Issues or PRs are very welcome.