Human Activity Recognition of Continuous Data Using Hidden Markov Models and the Aspect of Including Discrete Data

The combination of discrete, continuous data during complex activity recognition is presented, as well as a concept to analyze continuous data. Accelerometer, gyrometer data gathered from a body worn sensor are analyzed by a continuous Hidden Markov Model (cHMM). This cHMM is evaluated through two comparative studies, producing better, comparable results. The recorded data differ from other datasets because of more complex activities, leading to a more realistic environment representation. The complicated part of this task are differences in single activities. On the one hand order diversity of the subactivities, on the other hand similar activities in one class, like food preparation include preparing breakfast, a three-course menu. The final output gives a hint to use body worn sensors in combination with binary sensors.