Year of Graduation
Symbol Recognition in Kinetic Sensor Data Obtained from Mobile Phones
In this thesis we analyze acceleration data acquired from mobile phone sensors in order to recognize letters of Latin alphabet. This let us to enter textual data into the device using phone motions. There were developed a two stage system. The first stage is a basic gesture recognizer model, which could predict ”building blocks” of a letter. The second stage uses this prediction to classify a letter as a sequence of the basic gestures. In addition, there were compared different approaches to tackle the problem of sequential data classification such as LSTM and HMMs. The experiments with different configurations showed LSTM to be more robust and accurate for the both stages. Finally, we implemented and tested the system and discussed the results.