Explorations and Testing Ideas – Brain Computer Interface w/ Python, OpenBCI, and EEG data p.2

Running through where I am currently at, what I’ve learned, and what I’ve had trouble with while using the OpenBCI headset for reading EEG data.

https://openbci.com/

Channel membership: https://www.youtube.com/channel/UCfzlCWGWYyIQ0aLC5w48gBQ/join
Discord: https://discord.gg/sentdex
Support the content: https://pythonprogramming.net/support-donate/
Twitter: https://twitter.com/sentdex
Instagram: https://instagram.com/sentdex
Facebook: https://www.facebook.com/pythonprogramming.net/
Twitch: https://www.twitch.tv/sentdex

(Visited 2 times, 1 visits today)

Related Videos

Comment (21)

  1. Have you tried this experiment with your friends and family to see if there is a difference in the way their brains 'think' a class label? That'd be interesting if you can somehow quantify a similarity or dissimilarity measure between persons and create a relative scoring.

  2. Have you thought about using more differentiating thoughts? What comes to mind is a scene from Harry Potter where he must think of an exceptionally happy thought to conjure a Patronus. Thinking "left" or "right" literally might be too close to one another in he frequency domain to reliably measure. Therefore I postulate using drastically different thoughts to MEAN "left" or "right" may lead to better results. (i.e. Think about a really cringy moment to go left and think about a really proud moment to go right). I expect that using thoughts that light up totally separate parts of the brain is the way to go.

  3. instead of just thinking left try visualizing something that represents left. Like a road sign with a left turn at the same time as you are thinking left or internally repeating left? Our internal model is very visually driven.

  4. Ha! I almost mentioned Nyquist theorem in my last video when you were talking about the sample rate you need. In general, you actually want to sample much faster than the Nyquist rate to have a good representation of your signal. There is actually another equation that describes your measurement error based on your oversampling rate (I forget what that equation is called right now). The general rule of thumb we use in our industry is 10X oversampling rate, and anything less than 5X oversampling is regarded as too inaccurate for measurement purposes. If you look at the jupyter notebook code that you showed in the video, they are also only using data up to ~fs/6, or 6X oversampling ratio.You can see this in the frequency band limit they defined for gamma waves.

    Long story short, your system with 16 channels at 125Hz is really only good up to frequencies of ~25Hz. I think that in order to train your neural network to understand motor skill movements then you need a higher sample rate. You also should not be using fft data above fs/5.

  5. One of the key pieces of info to using this data well is understanding what each of the frequency ranges represent. Focus on the 15-20 hz range (beta waves) — this will represent conscious intention better than the other EEG frequencies

LEAVE YOUR COMMENT

Your email address will not be published. Required fields are marked *