Minimizing the Set Up for ADL Monitoring through DTW Hierarchical Classification on Accelerometer Data

Systems for remote monitoring of motor activities in the elderly are becoming very popular in developed countries. In this context, recognition and classification of Activities of Daily Living (ADL) is a very important step that can open intriguing scenarios, especially if real-time techniques become available. The present work proposes a hierarchical classifier based on the Dynamic Time Warping (DTW) technique, applied on data recorded from a tri-axial accelerometer placed on the shin, to classify among different motor activities. The classifier was applied to the recognition of walking, climbing and descending stairs of five different subjects. After the calibration phase needed to extract the templates, the technique makes it possible to recognize activities by determining the distance between the signal input and a set of the previously defined templates. Signals coming from the three different channels are used in a hierarchical way, with three layers. The hierarchy has been set based on sorting channels by signal to noise ratio in descending order. The results show a classification with overall percentage of error less than 5%.