Deep Learning project to classify and predict the keystrokes typed by user wearing smartwatches
Wearable consumer electronics with a myriad of sensors are becoming increasingly popular. With the boost in the popularity of smartwatches, the expectations from these de- vices are also on the rise. People desire more features than only counting steps. This project is aimed at a novel such use of a smartwatch-like device: to detect keystrokes typed on a classic QWERTY keyboard. Each key that one types is associated with a characteristic hand motion and fre- quency. Using time-series data provided by sensors inside the smartwatches worn on each hand of a user, we recover what is being typed on the keyboard. Since the data is af- fected only by the motion of one’s hands, we show that a deep network learns to model these hand motions correctly. We hope to be able to recover the intended words typed by a user imitating typing on a keyboard in places where a phys- ical keyboard is not readily available, thus reducing the re- liance on a physical keyboard in one’s daily life.