Construction worker's awkward posture recognition through supervised motion tensor decomposition

Awkward postures in construction activities pose substantial hazards in both instantaneous injuries and long-term work-related musculoskeletal disorders (WMSDs). Posture recognition using motion capturing systems shows promising potential in avoiding and minimizing workers' exposure to awkward postures. However, current motion capturing systems require huge computational resources and complicated processes to recognize postures in construction tasks. To address this issue, we proposed an abstract and efficient motion tensor decomposition approach to compress and reorganize the motion data. Together with a multi-classification algorithm, the proposed approach is able to efficiently and accurately differentiate various postures. To validate the approach, we employed a system based on inertial measurement units (IMUs) to examine two sample activities composed of sequencing postures. The results indicate the proposed approach is able to provide sufficient recognition accuracy with less computation power and memory. Also, the idea of tensorization and tensor decomposition in this paper is extendable to other studies in the construction industry.