[ New messages · Members · Forum rules · Search · Login ]  
Page 1 of 11
Forum » Confederation News Agency » Confederation Board » Electromyographic Gesture Recognition Algorithms! (2nd Place)
Electromyographic Gesture Recognition Algorithms!
remoDate: Sunday, 2016/January/31, 2:27 PM | Message # 1
Thursday, February 12th, 2015
Group: Administrators
Messages: 1010
Status: Offline


This weekend we had a 36 hour hack-a-thon event at my college. I lead a group that won second place!

We did a project taking electromyographic sensor data from a Thalamic Labs Myo and writing our own gesture recognition algorithms.

We captured data from the device using a program I had written previously and that we modified for the event to log to files:



Then we use some calculus and python code to detected when and over what interval a gesture has occurred:



Then we integrate each channel over the detected interval:









We captured three attempts for each of the seven test subjects for six different gestures.

Once we had our 8 scalar values for each gesture event, we converted them into eight 2 dimensional vector values by multiplying each scalar by the appropriate unit vector for a given sensor.

From those vectors we calculated values for the magnitude of the polarity and also overall oblateness.





Using our computed values for polarity and oblateness we were able to distinguish between fist based gestures and finger based gestures.

All of our source code, experimental data and plots can be found at the github repo below:

https://github.com/remoford/GestureAnalyzer

Yay 2nd place!
 
blindflashDate: Sunday, 2016/January/31, 2:49 PM | Message # 2
Forum Private
Group: Graduate
Messages: 5
Status: Offline
Good job man!

ayy lmao
 
Mikm0nkDate: Sunday, 2016/January/31, 2:53 PM | Message # 3
Forum Lieutenant
Group: Forum Visitor
Messages: 69
Status: Offline
PogChamp
 
NeonFireDate: Thursday, 2016/February/04, 9:52 PM | Message # 4
Thursday, February 12th, 2015
Group: Blocked
Messages: 526
Status: Offline
just remember you know nothing about infinity and theres no homeless people in britain

-Former CDA
-Not a communist
-LOOK AT MY SUIT
 
shortlandDate: Thursday, 2016/February/04, 11:15 PM | Message # 5
Forum Major general
Group: High Command
Messages: 387
Status: Offline
I miss u @NeonFire


[Captain] : [Admiral] : [Domestic Advisor]
 
remoDate: Sunday, 2016/March/20, 2:24 AM | Message # 6
Thursday, February 12th, 2015
Group: Administrators
Messages: 1010
Status: Offline
I've been less than active with the clan over the last month and a half as I have been working on an extracurricular research project based on the 36hr HackMT project. Today I reached something of a milestone and I'd like to share. Since it is 2am Sunday morning, instead of doing a big writeup I'll simply quote the email I sent to the participants in the project.



Guys, when we presented the project at HackMT we had some rough code, some data, some excel plots put together by hand, a bunch of handwritten explanatory plots and frankly, a lot of handwaving. What we did not have were good plots our of computed values exhibiting per gesture clustering from which we could do real gesture detection.

I've been continuing to work on the project ever since in fits and starts and I'm pleased to announce the first real results from the project:

https://cs.mtsu.edu/~jrf2m/sums.html

I won't go over the method in detail, it is more or less as you remember it from HackMT on a theoretical level.

The end product after interval detection, integration and vectorization are two scalar values for each detected gesture event. These are the angle of the sum of the vectors and a measure of oblateness.

You can see a scatter plot of oblateness vs angle on the third row of plots. Each type of gesture gets its own plot and each participant gets his own color coding.

Three of our gestures show clear, and for the most part mutually exclusive, clustering:
counterclockwise
fist
peace

Two of our gestures fail to show meaningful clustering:
clockwise
horns

One gesture shows clustering but its less clear and has a degree of overlap with our three good gestures:
thumbsup

There is a tremendous amount of refinement that can be done with the code to improve on these results and I think we would also collect some better controlled data. However, the general method seems to be valid. I am quite pleased with these initial results.

I want to thank each of you who participated in the HackMT project, this would not be possible without your ideas and data samples.

The project is far from over, but I think this is a natural place to stop, reflect and plan. I will probably want to collect a second data set at some point and I would appreciate any suggestions or code contributions you might have. I will be in touch as the project continues.

Thanks,
Remo
 
shortlandDate: Sunday, 2016/March/20, 3:40 AM | Message # 7
Forum Major general
Group: High Command
Messages: 387
Status: Offline
sounds awesome biggrin and complicated t.t


[Captain] : [Admiral] : [Domestic Advisor]
 
Forum » Confederation News Agency » Confederation Board » Electromyographic Gesture Recognition Algorithms! (2nd Place)
Page 1 of 11
Search: