This is a data science project that applies an ASL translation filter over a virtual camera which can then be used in teleconferences such as zoom, discord and skype. It tracks landmarks on your body such as your face, hands and pose, feeding the data into a machine learning model to make a prediction on what word is being signed.
Built for iNTUition v8 hackathon https://devpost.com/software/unmuted
Download OBS Studio
Click "Start Virtual Camera" (bottom right), then "Stop Virtual Camera"
Close OBS
- Download repo
git clone https://github.com/jerremyng/asl-filter.git
- Install dependencies
pip install -r requirements.txt
- Run python script
python deploy_filter.py
- Then simply select the virtual camera from your video-calling app (zoom,skype,discord)
The list of 6 words it can understand are found in the label_map.csv file ('afternoon', 'bye', 'fine','good','hello','morning')
To test these out, one can find the respective sign actions on google.