Self-calibration of sensor misplacement based on motion signatures.

Human motion monitoring with body worn sensors is becoming increasingly important in health and wellness. However, achieving a robust recognition of physical activities or gestures despite variability in sensor placement is important for the real-world deployment of body sensor networks. A novel self-calibration process of sensor misplacement based on repetitive motion signatures is proposed. A rotation matrix model is introduced to represent the impact of sensor misorientation. Dynamic time warping (DTW) is employed for choosing and synchronizing training and testing datasets. The information from repetitive motion signatures is then used to calibrate sensor misplacement. In this work, walking was used as an example of a motion signature that provides information for sensor misplacement calibration. To investigate the validity of this method, a large dataset of 57 walking traces over seven different subjects was collected. With the proposed algorithm, we show that in the lower body motion tracking experiment, step-length-measurement accuracy can be improved from 45.84% to 94.51%.