Comparison of median frequency between traditional and functional sensor placements during activity monitoring

Abstract

Long-term monitoring is of great clinical relevance. Accelerometers are often used to provide information about activities of daily living. The median frequency (f[subscript m]) of acceleration has recently been suggested as a powerful parameter for activity recognition. However, compliance issues arise when people need to integrate activity recognition sensors into their daily lives. More functional placements should provide higher levels of conformity, but may also affect the quality and generalizability of the signals. How f[subscript m] changes as a result of a more functional sensor placement remains unclear. This study investigates the agreement in f[subscript m] for a sensor placed on the back with one in the pocket across a range of daily activities. The translational and gravitational accelerations are also computed to determine if the accelerometer should be fused with additional sensors to improve agreement. Twelve subjects were tested over four tasks and only the “vertical” x-axis showed a moderate agreement (Intraclass Correlation Coefficient of 0.54) after correction for outliers. Generalizability across traditional and functional sensor locations might therefore be limited. Differentiation of the signal into a translational and gravitational component decreased the level of agreement further, suggesting that combined information streams are more robust to changing locations then singular data streams. Integrating multiple sensor modalities to obtain specific components is unlikely to improve agreement across sensor locations. More research is needed to explore measurement signals of more user friendly sensor configurations that will lead to a greater clinical acceptance of body worn sensor systems.