Assistant Professor of Biomedical Engineering Fairfield University, United States
Introduction:: Current marker-based "gold standard" approaches for motion tracking human biomechanics find exceptional accuracy, but they face multiple limitations. Marker-based systems are very expensive and require extensive training before they can be used. Infrared cameras can experience inaccurate or noisy measurements due to uncontrolled lighting conditions, which binds experiments to an indoor dedicated lab space. Subject set-up and calibration with these markers can also be time consuming, preventing high throughput capture of multiple subjects. Albeit the accuracy of gold standard approaches, they are not ideal in terms of accessibility, affordability, and overall convenience. Therefore, cost effective and open-source markerless approaches using computer vision and machine learning must be investigated in order to increase accessibility to human motion capture. The main question regarding these systems, and the questions we plan to answer, is are they as accurate in tracking human musculoskeletal movement as the gold standard marker-based systems?
Materials and Methods:: This study will examine 15-20 subjects' lower body kinematics while performing bodyweight movements including squats, countermovement jumps, heel raises, and 45 degree cuts. Marker clusters will be placed on the subject's sacrum, thighs, shanks, and feet, for a total of 7 marker clusters. The marker clusters will be tracked by the marker-based motion tracking system in the Fairfield University Biomechanics lab, using 8 infrared cameras (Optitrack and MotionMonitor). 3-5 low-cost cameras will also be set up on tripods for the markerless system. Both systems will record the subjects' movements simultaneously. Data between the two systems will be synchronized using an Arduino circuit powering a flashing LED. After synchronization, the lower body kinematic data between the marker-based and markerless systems will be compared using python to quantify the accuracy of the markerless system. This quantification will be done using the coefficient of multiple correlation (CMC), in which the "actual" values are the gold standard marker-based system's and the predicted values are the markerless system's. The first system we plan to validated is The FreeMocap Project, created by Jonathan Mathis, Assistant Professor of Human Movement Neuroscience at Northeastern University. This system works with 3 USB webcams and a print-out of a charuco (checker) board for calibration. While marker-based systems cost thousands of dollars, the total cost for materials (3 USB webcams, 3 tripods, 2 USB extenders) for this open-source software costs under $100.
Results, Conclusions, and Discussions:: Since this validation study involves human subjects, an IRB has been submitted and we are currently awaiting approval. Literature review and experimental procedures have already been put in place to ensure that data collection can begin immediately after IRB approval. This section will be updated once data collection has been performed and results have been evaluated.
