The blue bar

Working with OpenViBE signal processing scenarios and doing scenario/BCI design
Post Reply
rencarcos
Posts: 4
Joined: Tue Oct 25, 2011 2:24 pm

The blue bar

Post by rencarcos »

Hey people! How you doin'?
We need a little help here. We're using a scenario of BCI motor imagery with CSP at Ubuntu 11.04. OpenVibe 0.11.0.
At online scenario when we run the application it shows a blue bar that always stays at the right side even though we try to think to other side.
We think that this bar should show the side we are thinking of. Is this correct? Or this should get us only the hit rate? Or something else...
Help us! Please! What is the meaning of the blue bar? :shock:

jlegeny
Posts: 239
Joined: Tue Nov 02, 2010 8:51 am
Location: Mensia Technologies Paris FR
Contact:

Re: The blue bar

Post by jlegeny »

Hello rencarcos,

The blue bar is showing the mental state of the person, like you expected. If the bar always stays on one side (does it move at all?) then it means that there was a problem during the calibration step.

In order to help you, we will need more information about what you did before running the online scenario. How did you perform the training, which EEG hardware are you using and such.

Thank you
Jozef

rencarcos
Posts: 4
Joined: Tue Oct 25, 2011 2:24 pm

Re: The blue bar

Post by rencarcos »

Hi! Thanks for the reply!

We're using the EmotivEpoc. For the acquisition, training and online scenarios we are using the scenarios for motor imagery with CSP that comes with Openvibe.

The Bar stays always on the right side of the cross. It moves. Sometimes it's near the center of the cross(always on the right side) and sometimes it goes to the extreme right side of the window but never goes to the left side of the cross.

At the end of training scenario, it says that the classifier performance on whole set is about 80%.
When I use the online scenario with a Classifier Accuracy Measure box and changing the acquisition client box for a Generic Stream reader box reading the same .ov that i used in the training, it shows about 6%.

Thanks for the help

lbonnet
Site Admin
Posts: 417
Joined: Wed Oct 07, 2009 12:11 pm

Re: The blue bar

Post by lbonnet »

Hi rencarcos,

Just to be sure, here is the procedure :
- Run the scenario 0-signal-monitoring to check the signal acquisition and quality. You should be able to see for example the blinks as small spikes on the frontal electrodes.
- Run the scenario 1-acquisition to acquire some training data. It should last around 8 minutes if I remember well (20 trials Left, 20 trials Right, instruction as an arrow).
- Train the CSP spatial filters with the scenario 2-train-CSP. It should end with a "success" message.
- Train the LDA classifier with the scenario 3-train-LDA. It should end with the score estimation.
- Run the scenario 4-online. The blue bar moves according to the motor imagery of the hands. If in your case the blue bar is stuck on one side, and cannot move on the other side, it means the LDA is totally in the dark... classifying always in the same direction.

Now some serious warnings:
- The Emotiv EPOC is not using the good electrode sites for motor imagery. The motor cortices are roughly around C3 and C4, the EPOC has most of its electrode on frontal position.
- The training phase is crucial : the user must perform continuous mental imagery of a hand movement during each trial, according to the direction of the arrow.
- If the experimental conditions are not "good", you cannot trust the classifier accuracy after LDA training. The classifier may be for example over-fitted, being too much specialized on the training set.
- Motor imagery is a very difficult paradigm for naive users. It requires training, even with a medical-grade amplifier and many well-placed electrodes.

And finally some advices:
- Be sure to have good contact with the electrodes. The EPOC signal is quiet noisy...
- Try to place the headset on a non-conventional position, by trying to put more electrodes above the motor cortices.
- Try first with real hand movements, it's much easier to understand for the user. Then if you manage to have good detection of the movements, try with imagined movements.
- The training set is critical: the user must maintain the movement during the trial, and must remain still and relaxed.

I managed to get this BCI working with the EPOC, with very limited success on detecting the movements though... around 70% top.
for the record, it was some times ago and I didn't intensively test it...

Hope this helps

Laurent
Follow us on twitter >> openvibebci

Checkout my (old) blog for some OpenViBE tips & tricks : here !

kiyarash
Posts: 27
Joined: Tue Apr 11, 2017 10:44 am

Re: The blue bar

Post by kiyarash »

rencarcos wrote:Hi! Thanks for the reply!

We're using the EmotivEpoc. For the acquisition, training and online scenarios we are using the scenarios for motor imagery with CSP that comes with Openvibe.

The Bar stays always on the right side of the cross. It moves. Sometimes it's near the center of the cross(always on the right side) and sometimes it goes to the extreme right side of the window but never goes to the left side of the cross.

At the end of training scenario, it says that the classifier performance on whole set is about 80%.
When I use the online scenario with a Classifier Accuracy Measure box and changing the acquisition client box for a Generic Stream reader box reading the same .ov that i used in the training, it shows about 6%.

Thanks for the help
Hi I have the same low online accuracy as you. Were you eventually able to get a good online accuracy using the emotive epoc+ in the MI CSP scenario ?

if yes would you share your experience. (how did you tilt the headset or any other change that made things better)

Best regards,
Kiarash

Post Reply