openvino_face_detection

first-person face detection

CoverVideo2.mp4

Prints your face detection and name in your webcam stream. Possibility to hide the bounding box and name imprint.

Additional installation

1. Make sure your in the MeetingCam root directory

2. Download and convert a face detection model, then opt out from sending usage data

   omz_downloader --name ultra-lightweight-face-detection-rfb-320 --output_dir src/meetingcam/models
   omz_converter --name ultra-lightweight-face-detection-rfb-320 --download_dir src/meetingcam/models --output_dir src/meetingcam/models --precision=FP16
   opt_in_out --opt_out

   More about the used face detection model can be found here. This is a very light weight model, which will just work fine due to the easy task of face detection of a person in front of a webcam.

Usage

First add a camera device to MeetingCam

To show a modified camera stream in online meetings it is necessary to create a virtual camera for each real camera you want to use.

1. Activate the virtual environment

   source .venv/bin/activate

2. Get command to add a camera device

   python src/meetingcam/main.py add-devices

   Then copy and execute the command of the camera you want to use.
3. You can now see that the camera of your choice has a virtual counterpart.

   python src/meetingcam/main.py list-devices

Run a plugin

Replace yourname and device_path with your name of choice and your camera path which has a virtual counterpart.

python src/meetingcam/main.py face-detector --name yourname /dev/video0

Get help

Options and usage for the plugin are documented in the plugins help function.

python src/meetingcam/main.py face-detector

Showcase

By default the bounding box and name imprint are shown.

You can hide the name imprint by pressing <Ctrl+Alt+n>.

Also the bounding box can be hidden <Ctrl+Alt+f>.

Implementation details

The model itself is a general face detector, which means that it is meant to detect all faces in the webcam stream. To avoid additional bounding boxes on other faces the following assumption is made.

Assumption:
You as user are nearest to the camera and hence the detected bounding box of your face covers a bigger area then from faces detected in the background.

So this plugin is is just printing a bounding box on the face detection which covers the biggest area in the video stream. That means if a person passes through behind you, the bounding box will be still shown on your face because you are nearer to the camera and hence your face detection covers the biggest area. Just be aware that the implementation fails if the above assumption is not valid in your case/setup.