Introduction:
We created an armband that used infrared (IR) diodes to capture nger movements from the the extensor digitorum communis muscle. From here, we used a machine-learning model to classify these signals, in order to create a real-time predictor. The idea was to shine infrared light on the user's upper forearm. As the user moved di erent ngers, muscles on the upper forearm would move, causing the IR light to di ract in di ering directions. This re ection was detected using photodiodes, digitized through an ADC, and then processed via machine learning in order to predict exactly which nger the user was moving. Using two channels, we were able to achieve almost perfect prediction accuracy when distinguishing between three ngers in real time. While a third channel was built, due to time constraints we were not able to acquire three channels of data. Given the time, we believe we would be able to accurately predict four di erent nger movements, and possibly the thumb (the thumb does not really seem to cause much muscle movement in the upper forearm).
Full report (pdf)