Eliminating motion artifacts from fabric-mounted wearable sensors

Sensors embedded into clothing for measuring human movement are becoming more widespread in research, with applications in clinical diagnostics or rehabilitation studies. A major issue with their use is the undesired effect of fabric motion artifacts corrupting movement signals. This paper presents a method for learning body movements, viewing the undesired motion as stochastic perturbations to the sensed motion, and utilising errors-in-variables models to eliminate these errors in the learning process. Experiments, both in simulation and with a physical fabric-mounted sensor, indicate improved prediction accuracy as compared to standard learning methods.