ABSTRACT
Chapter 4 shows that human activities can be effectively recognized with acceleration and physiological signals. However, such recognition and the performance analyses were carried out offline, in a server. This chapter focuses on the next step in HAR: the online implementation (i.e., providing real-time feedback) on smartphones. Such a task is not trivial in view of the energy, memory, and computational constraints present in these devices. In activity recognition applications, those limitations are particularly critical since they require data preprocessing, feature extraction, classification, and transmission of large amounts of raw data. Furthermore, to the best of our knowledge, available machine learning API’s such as WEKA [18] and JDM [11] are neither optimized nor fully functional under current mobile platforms. This fact accentuates the necessity for an efficient mobile library to evaluate machine learning algorithms and implement HAR systems in mobile devices.
