GestureAI

GestureAI is a RNN(Recurrent Nerural network) model which recognize hand-gestures drawing 5 figures(Circl, Rectangle, Triangle, Cross and the other). This dataset of hand-motion drawing 5 figures is sequences of 3-axis accelerations captured by iPhone. Example to implement RNN in Keras gets 90.8% accuracy by Cross-validation.

Demo

Trained Neural Network deployed on GestureAI-iOS, iOS 11 app using CoreML :

Get the Dataset

You can use direct links to download the dataset.

Labels

The dataset consists of 1,000 3-axis acceleration sequences of 5 gesture classes, which are defined by motions drawing 5 figures. We don't set a specific rule about stroke order for drawing a figure by hand.

Label	Description	Examples	Figure
0	Circle	200
1	Rectangle	200
2	Triangle	200
3	Cross	200
4	Other	200

Requirement

• Python (2.7+)
• numpy (1.12.1+)
• protobuf (3.1.0+)
• Keras (1.2.2)
• TensorFlow (1.2.1)
• Scikit-learn (0.15+)
• coremltools (0.6.3)

Install

$ virtualenv venv
$ source venv/bin/activate
$ git clone https://github.com/akimach/GestureAI.git
$ cd GestureAI
$ pip install -r requirements.txt

Examples

Try with Jupyter notebook!

• Loading datasets
• Tuning hyper-parameters with Grid Search
• Training RNN with Early-stopping

Licence

MIT

Author

Akimasa KIMURA