07 – YOLO, this AI is working for me part 1/2

To Summarise again :
I’m still working on a my music concept : playing music with no music instruments, by moving coloured stuff if front of a camera !
You don’t believe me ? Just read my previous posts 😀

Right now, for the technical part, I’m tracking coloured balls placed of chopsticks with simple algorithm filtering colours (with openCV)

And it’s working quite well !

However, I’m having trouble, right now, to track the fast moving colours because of my cheep camera :

This is why I wanted to bring out the big gun this time :

Training an artificial intelligence
to do the work for me

And the AI name I will use is YOLO (You Only Look Once)

Yolo is a family of Convolutional Neural Networks (CNN) that are very efficient to do some image or object recognition and classification.

What are Convolutional Neural Networks (CNNs) ?

“Basically a brain, but in a computer”

me

Indeed, CNNs are inspired by the architecture of the brain. Just like a neuron in the brain processes and transmits information throughout the body, artificial neurons in CNNs take inputs, processes them and sends the result as output.

We give an image to the neural network as an input, each pixels are converted into numbers that goes to the output while going trough intermediate neurones layers called the hidden layers. This is why CNN are parts of what we call deep learning, it’s because they use a big amount of neurone layers !

How does the network take decisions ?

Each neurones take a decision if it send the information they receive to the next neurones. This decision is based on a weight, which is a number stored in the neurone and is added to the input.
All the neurones weight together form our model, and is what make it smart or not.

And how does it learn ?

By changing the neurones weights with try and error !
If you want to detect a ball on an image, you will give him an image with a ball on it, he will try to guess if there is a ball on the image and where it his. If the model fail, he will adjust the right neurones weight to succeed next time !

And what about Yolo ?

The Yolo models are pre trained, that mean that all the weights are already set and that they are ready to be used to detect common things such as humans, dogs, bikes, cars and so on :

But this model will be really helpful for me who just want to detect coloured balls. I don’t want to train an entire model from scratch.

So Yolo I will use
I hope my computer will be powerful enough to detect the balls in real time…

See you in the next post to see how I did it and what are the results