Finger Counter:

A program that counts the number of fingers you're holding out.

The data was collected using the webcome and OpenCV, and the Classification model is a CNN inmplemented in keras with Tensorflow as backend.

Requirements

• pandas==0.23.0
• numpy==1.14.6
• matplotlib==3.0.2
• scikit-learn==0.19.1
• tensorflow==1.9.0
• Keras==2.2.2

Contents

• Finger Counter.ipynb - jupyter notebook used for Creating the data, Training and testing the model.
• gestures - Folder containing the training images
• checkpoint - Folder containing a preatrained model
• hand_emo - Folder containing

Procedure

OpenCV was used as hand detector, we created a database consisting of 2000 images of binary masks, with 5 classes going from 1 to 5 fingers. A keras model was used as a classifier. we train a simple model on 85% of the data and validate it on 15%.

Training and Performance

The model is trained with TensorFlow backend using a NVIDIA GeForce 950M for 14 epochs using batches of 50 images each. The training input are not augmented, Augmentation techniques could improve the performance of this model. A Testing performance of greater than 99% accuracy is achieved.

Model Learning loss and Accuracy is shown below

Demo

Demo of Finger Counter Can be found here. https://www.youtube.com/watch?v=a0CjgK5weOg&feature=youtu.be

Resources

• Emojinator https://github.com/akshaybahadur21/Emojinator
• CNN-HowManyFingers https://github.com/jaredvasquez/CNN-HowManyFingers

ToDo

• Add the Zero fingers case
• fine-tune ConvNet a bit more
• Collect more data
• use data augmentation
• use a CNN model for the detection phase