This is a simple approach to Gesture Recognition using Deep learning in tensorflow. This uses a combination of Convolutional Neural Networks & variant of Recurrent Neural Network for recognising the hand and the movement (gesture) made using it respectively.

The recognised gestures are then (planned to be) mapped to actual Desktop controls to allow Gesture Control. The client program uses simple API requests to the deployed flask server.

• Python3
• tensorflow
• Jupyter Notebook

1. $ pip install -r requirements

2. $ jupyter notebook

3. Download the original dataset from here.

4. Copy it onto your local machine or google drive.

   NOTE: This repository was originally run in a free Google Colab VM environment.

5. Run the Preprocessing notebook after tweaking the paths & removing google drive mount if running locally.

6. Run the Data Loading notebook to start working on it.

• Jupytertheme module used for saving myself from default light theme bleeding eye damage.

  pip install jupyterthemes jt -t monokai -nf latosans -nfs 13 -f source -fs 11 -T -N -kl

• Notebook extensions from contrib, jupyter_contrib_nbextensions module used to add certain IDE like functionality to Jupyter Notebook.

  pip install jupyter_contrib_nbextensions

  jupyter contrib nbextension install

  Add the following extensions (for convenience):

  1. Hinterland
  2. Snippets
  3. Split Cells Notebook
  4. Table of Content
  5. Collapsible Headings
  6. Autopep8

• Widgets ipywidgets enables dashboard like controllers

  pip install ipywidgets

  jupyter nbextension enable --py widgetsnbextension

This is a Work-in-Progress, will be updated once the model has been trained & tested.