A hierarchical classifier to monitor ADL through dynamic programming on dual-axis accelerometer data

The new focus on active ageing in developed countries renders more urgent the availability of remote monitoring for motor activities in the elderly. Recognition and classification of Activities of Daily Living in this context open intriguing scenarios especially if real-time techniques are available. The present work proposes a hierarchical classifier for activity recognition that uses only a dual axis accelerometer placed on the shin, and the Dynamic Time Warping (DTW) algorithm. The classifier was applied to the recognition of walking, climbing and descending stairs of five different subjects. The first part is a calibration phase, to obtain the template signals, and the second part recognizes activities by determining the distance between the signal input and a set of previously defined templates. The signals of the two channels will be used in a hierarchical way. The results show a classification with overall percentage of error less than 5%.