StaticGestureRecognition

Project from Applied Deep Learning 2020 - TUWien

The project consists in a static hand gesture recognizer that maps gestures to user-defined commands.

Requirements

Install dependencies

1. virtualenv venv. Note: the program was tested with python 3.7, you can choose an interpreter using for example virtualenv venv -p=/usr/bin/python3.7
2. Activate virtualenv: . venv/bin/activate
3. pip install -r requirements.txt

Get the model

1. Download released model
2. Copy model to: hand_classifier/models/model_final.hdf5

Or from the project's root directory run:

mkdir hand_classifier/models && cd hand_classifier/models && wget https://github.com/lucamoroz/StaticGestureRecognition/releases/download/0.9/model_final.hdf5

Run

To run on the webcam: python main.py

To run the application in debug mode (and see the video stream and prediction confidence) run python main.py --debug.

There are multiple options available, to see all of them run python main.py --help.

Change commands

You can change the commands executed by modifying the file commands.json, which associates a command to each gesture.

The commands are passed to the underlying system and executed.

Test

From the project root folder, run:

py.test

Train on your hands

There is a dedicated python file that can be used to retrain the classification layer to fir your dataset.

1. Collect your dataset. You can use the script datset/data_script.py to quickly add images to a datset, see dataset/README.md for more info.
2. Change the labels of hand_classifier/HandCNN.LABELS and commands.json according to your dataset classes.
3. Retrain the classification layer of the pretrained model: python hand_classifier/retrain_top.py --dataset [PATH_TO_DATASET] --model [PATH_TO_TRAINED_MODEL]