Gesture Classifier Possible Fits
Preprocessed emg data from the current hardware landscape with varied percent contractions across three subjects was taken and used to train a gesture classifier for software implementation. The three fits shown below were considered for use in the current scheme. The third order polynomial presented too much variation at the extremes of the data set (with the curves around 100% contraction). The piecewise linear fit was promising because it shows a valid data fit, but also would not take up too much processing power within the current Teensy hardware because of its simplicity. The other peicewise function also showed a good fit to the data set, but could complicate the processing due to the change in function across the different schemes. Moving forward with the linear data, we would have to account for differences across subjects with varying MVCs given muscle activity. This thresholding will be built into the software and automatically change the two extreme piecewise functions.
