I trained a hand sign detector using CNN (Convolutional Neural Networks) model in MATLAB. This was my final project for one of my subjects when I was pursuing my bachelor's degree. Below, you can find the full tutorial and source code I discovered on YouTube.
https://www.youtube.com/watch?v=P2oJVMMZXD8
https://www.youtube.com/watch?v=JU2qHAE0h2Q
Due to limitations on GitHub, I have placed the training dataset and the model (named MyNet.mat) on this Google Drive link.
https://drive.google.com/drive/u/0/folders/1TETWqbW1_QdehxAYFM6bRY5Zavwafj1g
1. run.mat
Execute this in your MATLAB, and it will create a training dataset based on your webcam. Make sure to place your hand inside the box while the program is running. I repeated this step for all 26 alphabets (A - Z) and created one blank sample for my identity (13319061 Riri Raissa).
2. test.mat
Once I obtained my dataset, I ran the training data using this module. I can customize the layers, but I chose the default ones from the tutorial.
g=alexnet;
layers=g.Layers;
layers(23)=fullyConnectedLayer(26);
layers(25)=classificationLayer;I used an alexnet (8 layers deep), 26 connected layers for the 26 alphabets, and a classification layer at the end to identify the displayed alphabet.
3.programtrue.mat
After obtaining the model named MyNet.mat (which I didn't upload here because it's over 600 MB), I can run the program to identify the displayed alphabet.
For futher explanation, you can see my video demo in here:
https://drive.google.com/file/d/1JVB644wSrpZrQWNSZ_MRfQSL4UeIf3ld/view
This the details of the model that i trained:
In the future, when i've super pc, I would like to recreate this project with TensorFlow in Python langauge. Maybe it will perform better at detecting hands instead of pictures.
Cheers to the author of AlexNet and Knowledge Amplifier on YouTube for their awesome tutorial videos!