top of page
Two methods have been used for data analysis
  • Pearson's Correlation Coefficient 
  • Dissimilarity Measure (inspired by Texture Synthesis Algorithm)
​
Myo Device Utilization
​
The Myo device has 8 EMG sensors(we number them 0-7) and can sense spatial and gestural data.The device itself has many other gesture recognition capabilities that has not been used in this project. Data from the 8 sensors have been collected for all experiments and verification.
​
Pearson's Correlation Coefficient:
​

Correlation : In order to check whether two graphs are similar Pearson’s correlation coefficient is computed both individually for the 8 sensors and also for the summed up value of the sensors.

The coefficient is value between 1 and -1. Positive 1 means that the two samples match and -1 means that they are completely opposite. The following image depicts the r-value calculation. 

​

​

Here x  and y are data points in Sample A and Sample B respectively and n is the number of data points used in the computation.

​

For this project, a naive windowing technique has been used that is similar to finding substring in a given string. For obvious reasons (like varying typing speed to imperfections), the samples collected from the same person typing the same thing may not generate an exact replica (not same amplitude and not through same time) of previous attempts.

​

Thus a windowing technique has been used to compare two data sets or samples. The technique simply requires to take a subsequence of consecutive data samples of specified size (window size) and compare against every such chunks(window) of data of the other samples.

​

​

There does exists discrepancies to the extent correlation works. We did observe that apart from the few initial tests correlation fails to classify properly. This might be caused due to :

​

1) Data not collected properly, there might be noise in the data. Not all users are touch typist.

2) Most sensors remain dormant as seen from the data, only half of the sensors were somewhat responsive with amplitude changes.

3) Not enough time spent in investigation other formulas, approximations and smoothing.

​

Measuring Dissimilarity

​

Steps:

  1. Selecting significant sensors

  2. Gaussian Blurring (Smoothing)

  3. Window Dissimilarity

  4. Gaussian Weighting

  5. Aggregate Dissimilarities

​

As previously explained a similar sliding window technique has also been used here. A dissimilarity by taking the difference between current windows of the two data sets in questions after some preliminary smoothing and noise cancellation is done.

​

​

​

bottom of page