Marker-less versus marker-based driven musculoskeletal models of the spine during static load-handling activities

Evaluation of workers' body posture in workstations is a prerequisite to estimate spinal loads and assess risk of injury for the subsequent design of preventive interventions. The Microsoft Kinect™ sensor is, in this regard, advantageous over the traditional skin-marker-based optical motion capture systems for being marker-less, portable, cost-effective, and easy-to-use in real workplaces. While several studies have demonstrated the validity/reliability of the Kinect for posture measurements especially during gait trials, its capability to adequately drive a detailed spine musculoskeletal model for injury risk assessments remains to be investigated. Lumbosacral (L5-S1) load predictions of a Kinect-driven and a gold-standard marker-based Vicon-driven musculoskeletal model were compared for various standing static load-handling activities at different heights/asymmetry angles/distances. Full body kinematics of eight individuals each performing eighteen activities were simultaneously recorded by a single-front-placed Kinect and a 10-camera Vicon motion capture system and input to AnyBody Modeling System. The predicted spinal loads by the two models were in average different by 17.8 and 25.9% for the L5-S1 disc compressive and shear forces, respectively, with smaller errors for the activities at higher load heights. Some activities performed near the floor could, however, not be recorded by a single-front-placed Kinect sensor due to the joint occlusion. The capability of the Kinect to adequately drive a spine musculoskeletal model depended on the complexity of the activity. While a single front-placed Kinect camera can be used to evaluate spinal loads in a wide range of static/quasi-static activities, cautious should be exercised when evaluating tasks performed near the floor.