Automatic pattern recognition of functional upper-limb activities using Markov chains

Abstract

The development of algorithms based on Markov’s chains for automatic recognition of upper-limb basic activities of daily living (ADLs) is presented in this paper. Such activities include, among others; combing hair, eating with a spoon, drinking from a glass and tooth brushing. A kinematic signal was used to feed the recognition system. This signal was obtained using an electro-hydraulic activity sensor on each upper limb. This sensor produces a signal that describes vertical position of wrist relative to shoulder. Data were obtained of 7 volunteers. The parameters used in the automatic recognition process were: classification of movement, no-movement and relative position of wrist in functional-vertical intervals. The later was used as states in the Markov’s chains models. Two models were designed for automatic recognition; Markov’s chains and a hybrid one. The hybrid system was a combination of Markov’s chains and decision trees. The developed algorithm reached up to 96.7% of accuracy for some tasks. This system can be used, as a complement of an ambulatory monitoring system, to detect functional upper-limb activities during the whole day.