Kinematic Model Based Sensor Fusion for Inertial Measurement Units in Injury Biomechanics
Loading...
Date
2021-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Accurate kinematic measurement of human body segments is central to many investigations in biomechanics. The use of Inertial Measurement Units (IMUs) for this purpose is becoming increasingly effective in low acceleration regimes in part due to the incorporation of human body kinematic models in the sensor fusion process. For impact scenarios in injury biomechanics, the required sample rates and measuring ranges are larger than those of current commercial IMU systems with human model based fusion algorithms. Without sophisticated fusion algorithms that interface with impact-rated IMUs, injury biomechanics research has to rely on camera-based motion capture systems to ensure reliable measurements of body segment position and orientation (i.e., pose). The objective of this research is to develop a novel model based sensor fusion algorithm for impact-rated IMUs to reduce the necessity of camera-based pose measurements in injury biomechanics experiments. The scope of the present study focuses on state estimation for anthropomorphic test device (ATD) upper limbs, but the methodology can be intuitively extended to the entire body. The algorithm takes in acceleration, angular velocity, and angular acceleration measurements from IMUs mounted to the ATD and outputs piecewise polynomial estimates for the joint angles between ATD segments, as well as the six degrees of freedom of the ATD thoracic spine with respect to the inertial reference frame. These polynomials can be differentiated analytically to evaluate the full kinematic state of the ATD upper limb/thoracic spine assembly. The algorithm uses offline optimization for maximum likelihood estimation, relying on a zero mean gaussian noise model for the measurements and a kinematic model for the ATD upper limb/thoracic spine to evaluate the likelihood of a given set of polynomial estimates. For simulated measurements augmented with gaussian noise based on the sensor uncertainties, the algorithm was promising. Due to restrictions on sensor availability, experimental validation of the algorithm was postponed, but future work will use physical experiments to compare the algorithm pose estimates to measurements from a camera-based motion capture system.