Activity classification using a state transition diagram and activity levels

Wireless health monitoring systems are suggested for use in the health care of the elderly to enable independent but secure living for as long as possible. Activity monitoring is highly valued by welfare and health personnel because it gives an overview of the well-being of the elderly. In this paper, a state transition diagram based activity classification MATLAB-algorithm is described. The state transition diagram takes consideration of the reasonable sequential states. The system distinguishes between different states and activity levels with the help of a single waist worn measurement unit consisting of a tri-axial accelerometer and a tri-axial gyroscope. The allowable states are stand, sit, lie, walk, and fall. Every second of data is classified into one of the previous states, but also into one of the four possible activity levels: passivity, light activity, activity, and intense activity. The performance of the state classification is tested with the data collected in a home environment. The sensitivity and specificity values for the state transitions are in the range of 80-100%. The classifier detects longer continuous periods in the same state with a minimum probability of 85.7%.