| Resumo: | Built-in hardware sensors in many of the modern smartphones, such as accelerometers and gyroscopes, open a world of infinite opportunities for novel applications based on the context perceived from the data they provide. Human activity recognition (HAR) is a direct application of this technology, which despite being a very active field of study in the past years, leaves many strategies left to explore and key aspects left to address. A commonly ignored challenge of HAR is the difference of input signals produced by different people when doing the same activities. As a result, the activity classification method should be able to generate adapted results for each different user. This document proposes and explores a solution to this problem by means of "Online Semi-supervised Learning", an underexplored incremental approach capable of adapting the classification model to the user of the application by continuously updating it as the data from the user's own specific input signals arrives. The ideal scenario of this project would be the creation of a smartphone application capable from the beginning of classifying the user's activities with a certain error, and as the time passes and the user utilizes the application, without manual input, the system's classification error would decrease autonomously until it is virtually insignificant for that specific user. Several classification models will be generated from different online semi-supervised approaches, and further evaluated and compared, in order to decide on a best fit. The success of this approach would result in innumerable applications, and could considerably enhance the current interaction between people and their mobile devices, taking the concept of "smartphone" to a whole new level.
|
|---|