NNiimote update

Tuesday, July 10. 2007

This is a video with a small demo for the NNiimote.

NNiimote is just a small proof of concept appplication that uses a neural network to recognize Wiimote gestures. The neural network was implemented in R (I didn't actually implement the net, I used the AMORE library) and the "driver" program was made in Python. The Wiimote data is feed to the Python program via OSC from GlovePIE and the Python program invokes the net via the Rpy package.

In the video you can see it recognizing the three moves that the network was trained for:

Yes moves
No moves
Attack moves

One movement is "delimited" by pressing the A button on the Wiimote.

If you wish to play with this you can download the source at http://jorgecardoso.eu/MSc/NNiimote/NNiimote.zip. I'll upload the zip when my FTP connection gets restored.

Posted by Jorge Cardoso in Wii at 17:39 | Comments (7) | Trackbacks (0)

Trackbacks

Trackback specific URI for this entry

No Trackbacks

Comments

Display comments as (Linear | Threaded)

I notice that the AMORE library is a normal feed forward NN with backpropagation. How did you handle the temporal sequence of data i.e. what features did you use as inputs to the neural network?

#1 Gerrit Niezen on 2008-02-22 09:41 (Reply)

Hi, I used the acceleration data (X, Y, Z) from a set of sequential readings of a 'gesture'. A gesture was initiated and terminated by pressing a button. The set was condensed to 150 interpolated values (3*50) and fed to the NN. But NNiimote was just an exercise... I think that a gesture recognition using regular expression (http://silentlycrashing.net/ezgestures/) is probably better for simple gestures...

#1.1 Jorge Cardoso (Homepage) on 2008-02-22 11:22 (Reply)

are you able to provide the glovepie script that you wrote for nniimote too?

#1.1.1 Glycemia on 2008-04-22 05:48 (Reply)

Sure, I've put it in http://jorgecardoso.eu/MSc/NNiimote/WiiOSC.PIE

#1.1.1.1 Jorge Cardoso (Homepage) on 2008-04-22 09:43 (Reply)

Thanks for the PIE file. I've taken a look at the R file and noticed that I need 5 data files: yes-moves, no-moves, attack-moves,negative-moves and train-moves. May i ask what are these files specifically? Does it mean that i have to use the nniimote-save-moves to train the yes,no,attack moves? if so, what are the negative-moves and train-moves? And do i have to put in 150 readings for each data file?

#2 Glycemia on 2008-04-22 09:57 (Reply)

Hi, I think those "train.moves" aren't used... It's probably some leftover from my experiences. The 'Negative-moves' are moves diferent from all the rest. (An NN is probably not the best way to recognize moves because there is allways an extra class of moves which is impossible to define: my negative-moves are just an attempt to define them...) The train files are generated by the niimote-save-moves.py file. You should train the yes, no, attack and some other moves (negative).

#2.1 Jorge Cardoso (Homepage) on 2008-04-22 15:43 (Reply)

I have tried your nniimote with success! Thanx.... But i do have a few questions: 1) Why did you use 150 inputs? Was it arbitrary on your part? Coz i tried it with 90 and there was no significant difference in the LMS values obtained. 2)Why did you use the ADAPTgdwm method instead of the rest? 3) Why did you use 10 Hidden Layers in your ANN? 4) Why did set the momentum global to 0.5? Thanks in advance and sorry for bothering you with all these questions...

#3 Glycemia on 2008-04-26 08:56 (Reply)

Add Comment

Name
Email
Homepage
In reply to
Comment	BBCode format allowed Standard emoticons like :-) and ;-) are converted to images. E-Mail addresses will not be displayed and will only be used for E-Mail notifications To prevent automated Bots from commentspamming, please enter the string you see in the image below in the appropriate input box. Your comment will only be submitted if the strings match. Please ensure that your browser supports and accepts cookies, or your comment cannot be verified correctly. Enter the string from the spam-prevention image above: Enclosing asterisks marks text as bold (word), underscore are made via _word_.
	Remember Information? Subscribe to this entry