This experiment is a collection of "tools" to help create image data using the webcam in your browser. The video above shows how I trained a model to recognize me in front of the computer. Later, I'll may capture hand gestures and atribute meaning to it like: play a song or scroll the page. I do thing I'll need to make some improvements to my MLP code to make it more robust, but that'll come with time.
captureData.html - Get image data from the webcam.
on the label input, entry the label in the one-hot encoded hormat. e.g.: [1,0] if something is A - [0,1] if something is B.
How to use it
c - capture the current frame and add it to the data set.
s - save the data set (x, y) as x and y .json files.
u - undo. remove the last entry from the data set.
fitData.html - Train the neural network with the data created in captureData.html
testData.html - Test the neural network with new images from the webcam.
The nn.json model provided in this repo may not work for you, since it was only trained with 80 images of myself. 40 images of me in front of the computer, 40 of me not in front of the computer. The images ware scale to 32x24 pixels before training and during test.