A lightweight activity classification algorithm suitable for microcontrollers is presented, which is intended to be used on activity monitors. Therefore it focuses on five daily activities: inactivity, walking, cycling and walking up- and downstairs. This algorithm includes a novel approach for detecting cycling, which relies on properties of the power spectrum. Classification parameters are extracted from accelerometer and barometer data streams. Resources on activity monitors are usually short. The algorithm is able to cope with these limitations and was implemented on our demonstration activity monitor. Classification results were obtained from two test trials. The first trial consisted of consecutive sequences of basic activities. In the second trial a complex daily activity was executed. Twelve persons participated in each trial. Classification rates in the first trial were very promising: inactivity (97.3 walking a straight plane (92.6%), cycling (82.2%), walking upstairs (66.8%) and downstairs (65.7%). However trial two depicted that a supplementary class should be introduced, which requires further research.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below