3D human body reconstruction for worker ergonomic posture analysis with monocular video camera

In the modular construction industry of Canada, workers experience awkward postures and motions (reaching above shoulder, back bending backward, elbow/wrist flex, etc.) due to improper workstation designs. The awkward postures often lead to worker injuries and accidents, which do not only reduce the productivity but also increases the production cost. Therefore, the ergonomic posture analysis becomes essential to identify, mitigate and prevent the awkward postures of workers when workstation designs are changed. This paper proposes a novel framework to conduct the worker ergonomic posture analysis through the 3D reconstruction of human body from the video sequences captured by a monocular camera. The framework consists of four components: tracking worker of interest; detecting worker joints and body parts; refining 2D worker pose; and generating 3D human body model. The human body model generated from the framework could be used to estimate the joint angles of the workers to identify whether their postures meet the ergonomic requirements. The proposed framework has been tested on real construction videos, and the test results showed its effectiveness.