Evaluation of time-varying working posture based on interjoint coordination features extracted from sparse structure learning

Abstract

To evaluate tasks that have low physical workloads or small variations in the working postures, a method that can be used to analyze slight differences in postures and movements based on human motion characteristics is required. The interjoint coordination, which produces multijoint movements by organizing the redundant degrees of freedom into fewer major covarying relationships, is one of the primary characteristics of human motion. This work proposes a novel evaluation method for interjoint coordination using the graphical lasso and clarifies its efficacy for evaluations of interjoint coordination in complete tasks as well as time-varying interjoint coordination and working postures. In a subject experiment, eleven male participants performed lightweight material-handling tasks under different working conditions and paces, and an electromagnetic motion-tracking system was used to measure their working postures. The principal interjoint coordination for measured joint angles was extracted using the graphical lasso tool as a sparse precision matrix. Further, the contribution of each joint angle to the changes in the working postures in the time series was calculated by correlation anomaly using the graphical lasso. The estimated sparse precision matrix using the graphical lasso suggested that the principal interjoint coordination reflecting the differences in movement strategies by task type can be extracted in comparison with the covariance matrix. The time-varying correlation anomaly score suggested that the joint angle that contributes to differences between interjoint coordination at the onset and termination of tasks could be determined according to the final score. Therefore, our study shows the efficacy of the graphical lasso for evaluation of interjoint coordination and working postures from the time series for repetitive lightweight material-handling tasks.