Tylan Templin ^1* Christopher Riehm ^2,3,4 Travis D. Eliason ¹ Tessa Hulburt ^2,3,4 Samuel Kwak ^2,3,4 Omar Medjaouri ¹ David Chambers ¹ Manish Anand ^2,3,4 Kase Saylor ¹ Gregory Myer ^{2,3,4,5,6,7,8} Daniel Nicolella ¹

• ¹ Southwest Research Institute (SwRI), San Antonio, United States
• ² Sports Performance And Research Center, Department of Orthopaedics, School of Medicine, Emory University, 100 Woodruff Circle, Georgia, United States
• ³ Emory Sports Medicine Center, Atlanta, Georgia, United States
• ⁴ Department of Orthopaedics, School of Medicine, Emory University, Atlanta, Georgia, United States
• ⁵ Wallace H. Coulter Department of Biomedical Engineering, College of Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
• ⁶ The Micheli Center for Sports Injury Prevention, Waltham, Massachusetts, United States
• ⁷ Youth Physical Development Centre, Cardiff Metropolitan University, Cardiff, United Kingdom
• ⁸ Department of Mechanical Engineering, Indian Institute of Technology Madras, Chennai, Tamil Nadu, India

3D Markerless motion capture technologies have advanced significantly over the last few decades to overcome limitations of marker-based systems, which require significant cost, time, and specialization. As markerless motion capture technologies develop and mature, there is increasing demand from the biomechanics community to provide kinematic and kinetic data with similar levels of reliability and accuracy as current reference standard marker-based 3D motion capture methods. The purpose of this study was to evaluate how a novel markerless system trained with both hand-labeled and synthetic data compares to lower extremity kinematic and kinetic measurements from a reference marker-based system during the drop vertical jump (DVJ) task. Synchronized video data from multiple camera views was co-recorded with marker-based data from 127 participants who performed three repetitions of the DVJ. Root mean squared error values of lower limb joint angles and joint moments were ≤ 9.61 degrees and ≤ 0.23 Nm/kg, respectively. Pearson correlation values between markered and markerless systems were 0.67-0.98 hip, 0.45-0.99 knee and 0.06-0.99 ankle for joint kinematics. Likewise, Pearson correlation values were 0.73-0.90 hip, 0.61-0.95 knee and 0.74-0.95 ankle for joint kinetics. These results highlight the promising potential of markerless motion capture, particularly for measures of hip, knee and ankle rotations. Further research is needed to evaluate the viability of markerless ankle measures in the frontal plane to determine if differences in joint solvers are inducing unanticipated error.