A Tensor based Missing Samples Recovery for Human Movement Acquisition

Abstract

Monitoring of daily activities plays a crucial role in improving healthcare services and supporting clinical professionals. Wireless inertial measurement units (IMUs), which contain accelerometers, gyroscopes and optionally magnetometers, allow the acquisition of kinematic data outside of laboratory spaces. However, the acquired signals are prone to noise and missing values. In this work, a novel approach for missing data recovery from interdependencies between variables is proposed. Since multiple sensors are located in different parts of the human body and each sensor generates a multivariate signal, a tensor data structure (higher order matrix) is used with the ability to store multivariate spatiotemporal data. The proposed multivariate spatiotemporal tensor completion algorithm can effectively and efficiently recover data when a variable is missing temporarily or when one of the IMU sensors has completely failed to provide data. The proposed algorithm can be easily extended to find abnormalities or remove noise.