Accuracy Evaluation of Human Gait Estimation by a Sparse Set of Inertial Measurement Units

Abstract

Inertial measurement units (IMUs) have been utilized as motion capture (MoCap) devices in computer graphics, biomechanics, and rehabilitation. Typically, full body motions are estimated from the orientation and/or acceleration data of 13–17 IMUs attached to the body segments of experimental subjects. However, attaching numerous IMUs is quite intrusive and sometimes restricts the subjects’ motions. Recent advances in machine learning technologies have enabled full body motion estimation from a sparse set of IMUs (6–7 units). The present study compares the motion estimation accuracies of a system with a full set of IMUs (called full IMU MoCap) and a system with a sparse set of IMUs (called sparse IMU MoCap). Three male subjects performed three walking trials with different stride lengths (normal, short, and long), and their full body motions were estimated by each MoCap. Finally, the gait-related factors were calculated from each set of motion estimation results, and compared with the ground-truth data obtained by an optical marker-based MoCap. Although the sparse IMU MoCap achieved a lower overall accuracy than the full IMU MoCap, it can potentially evaluate the relative changes in the functionality of the locomotor during walking.