Recognition of activities of daily living based on the vertical displacement of the wrist

The upper limbs plays a key role in the performance of ADL, from support to total execution. However, Often rehabilitations therapies sacrifices the regaining of functionality to other functions as transportation. Therefore an automatic system capable of recognize ADL based on the information provided by upper limbs is needed. Lots of algorithms have been reported using inertial sensors with limited results. The aim of this work is to describe an algorithm to recognize some ADL performed with the upper limb, such as eating, drinking, talking by phone, combing hair and brushing teeth. The algorithm is based on an alternative novel sensor that provides information of the vertical displacement of the wrist relative to the shoulder, and can be used in a free-living environment. The detection system combines decision trees (DT) and Hidden Markov Models (HMM). Efficiencies reported goes from 61% up to 100%.