A Data-driven Functional PCA Filter for Compensating the Effect of Sensor Position Changes in Motion Data

Human body motion is typically captured by body area sensor networks. Accurate sensor placement with respect to anatomical landmarks is one of the main concerns in reliability of the motion capturing systems. Changes in position of sensors cause increased variability in motion data. Our goal is to isolate the characteristic features that represent the principle motion pattern. By using functional Principal Component Analysis (f-PCA) we compensate for the variation in data due to inadvertent movement of sensor placement. F-PCA is an effective tool for the study of human motion modeling by identifying hidden combinations and relationships between variables. The collected data from our experiment show differences between similar actions within different sessions of marker wearing. After applying f-PCA to the data, we show how the uncertainties due to sensor position changes can be compensated for.