ABSTRACT
As was mentioned in Section 4.1.2, feature extraction aims to filter the most relevant information within the collected signals and helps the classifier to effectively discover patterns within the data. In our context, measurements collected from the sensors are naturally indexed over the time dimension, which makes HAR a time-series classification problem. Nonetheless, the sampling rate of the sensors is not necessarily the same, thus, an accelerometer could provide observations at 100Hz whereas a heart monitor might deliver measurements at 1 Hz. This fact differentiates HAR from a typical multidimensional time series classification problem. The present chapter addresses the methodology followed by Vigilante to extract features from the measured signals in real time, including the data abstraction and the implementation of a number of well-known feature extraction procedures.
