CFE 2017: Start Registration
View Submission - CMStatistics
B0944
Title: Classification of human activity via spherical representation of accelerometry signal Authors:  Michal Kos - University of Wroclaw (Poland) [presenting]
Jaroslaw Harezlak - Indiana University School of Public Health-Bloomington (United States)
Malgorzata Bogdan - Lund University (Sweden)
Nancy Glynn - University of Pittsburgh (United States)
Abstract: Human health is strongly associated with person's lifestyle and level of physical activity. Therefore, characterization of daily human activity is an important task. The accelerometer is a wearable device which enables precise measurements of the acceleration changes over time of a body part to which it is attached. It can collect over a 1,000,000 observations per hour. The signal from an accelerometer can be used to classify different types of activity. We propose a novel procedure of activity classification, which is based on the spherical representation of the raw accelerometry data. Accurate classification information is obtained from the angular part of a signal, which is partially summarized via the spherical variance. One of the method's main properties is its ability to provide classification of short term activities. The classification accuracy of our method is 90\% for the within-subject level and 83\% for the between-subject level. In summary, the major contribution to the accurate classification of different physical activity kinds is the use of the spherical variance exhibiting rotational invariance. This property makes it insensitive to the shifts of an accelerometer location.