HandSignDetection-realtime

This is a live hand sign recognition program, based on the HaGRID dataset by hukenovs, which constitutes 18 distinct classes of hand signs.

I made this as a part of my Open-Ended project for my Generative AI class. It is based on Convolutional Neural Networks in Keras as well as some Image Preprocessing using OpenCV, Google's MediaPipe was used to extract relevant parts of images.

I achieved max accuracy score of 91.21% in model training and validation, although the no. of epochs were limited.

Screenshots

Dependencies

• Python
• Keras
• OpenCV
• MediaPipe
• MatplotLib
• Jupyter

Setup

To run this program, clone it to your local machine using:

git clone https://github.com/magnusjwatson2786/HandSIgnDetection-realtime.git

then cd to the repo directory and type:

python -m pip install -r requirements.txt --user

To download the dataset yourself, type:

wget https://n-ws-620xz-pd11.s3pd11.sbercloud.ru/b-ws-620xz-pd11-jux/hagrid/hagrid_dataset_new_554800/hagrid_dataset_512.zip -O hagrid_dataset_512.zip

or go to this link to download it manually.

To preprocess the data to make sure its ready for training:

python preprocessing.py

Usage

Now, to directly run the program using the pretrained model(make sure the program can access your camera):

python live_detect.py

However, if you wish to train the model yourself, use Training.ipynb python notebook using Jupyter.

License

MIT

Free Software, Hell Yeah!

Happy Googling!