Identifying the number and location of body worn sensors to accurately classify walking, transferring and sedentary activities
Omar Aziz, Stephen N. Robinovitch, Edward J. Park · 2016
In order to perform fall risk assessments using wearable inertial sensors in older adults in their natural settings where falls are likely to occur, a first step is to automatically segment and classify sensor signals of human movements into the known `activities of interest'. Sensor data from such activities can later be used through quantitative and qualitative analysis for differentiating fallers from non-fallers. In this study, ten young adults participated in experimental trials involving several variations of walking, transferring and sedentary activities. Data from tri-axial accelerometers and gyroscopes were used to classify the aforementioned three categories using a multiclass support vector machine algorithm. Our results showed 100% accuracy in distinguishing walking, transferring and sedentary activities using data from a three-sensor combination of sternum and both ankles.