Concurrent Evaluation of Marker-Based and Markerless Motion Capture for Assessing Sex-Based Differences in Lower Limb Kinematics and Anthropometrics

Abstract

Markerless (ML) motion capture has become increasingly popular in biomechanics because it addresses many practical limitations of marker-based (MB) systems by being less time consuming and more user friendly, and it enables movement data to be collected outside traditional laboratory environments. Systems such as Theia3D have been used to evaluate military and sports performance and in clinical assessments. However, while numerous validation studies have demonstrated promising accuracy, important limitations remain. These include limited transparency regarding kinematic model constraints and training datasets, as well as systematic differences between ML and traditional MB motion capture in joint centre definitions and allowable range of motion in each joint and anatomical plane. These differences make it challenging to distinguish true biomechanical differences from measurement-related differences. This consideration is particularly relevant when examining sex-based biomechanics, where known differences in movement patterns, injury mechanisms and musculoskeletal conditions exist. Understanding how ML and MB systems compare in capturing these differences is important for both research and clinical interpretation of results.

The purpose of this study was to compare MB and ML motion capture in their ability to detect established sex-based anthropometric and kinematic differences, and to determine whether the differences between motion capture systems are equivalent across sexes. Twenty-eight healthy adults (14 males, 14 females; 27 ± 4 years) completed self-selected-speed walking and drop vertical jumps (DVJ) while data were recorded simultaneously with both systems. To improve comparability, MB segment coordinate systems were aligned to the ML reference frame using a mean transformation matrix prior to kinematic analysis.

Both systems detected expected sex differences in hip width and shank length (p < 0.001). However, ML estimated notably shorter femur lengths compared to MB and did not identify a significant sex-based difference in thigh lengths, a well-established relationship in the literature. Comparison of joint centres revealed systematic offsets between the motion capture systems, with total mean offset of 3.37 ± 1.4 cm at the pelvis and 4.02 ± 1.3 cm at the hip. In general, ML pelvis and hip joint centres were positioned more distally compared to MB, across both static and dynamic trials.

For kinematics, both systems identified robust main effects of sex in frontal plane pelvis and hip motion. Sagittal plane waveforms showed very strong agreement between systems (r = 0.98). However, agreement was substantially weaker in the transverse plane for hip, knee, and ankle rotation (r < 0.5). Applying local coordinate system (LCS) alignment reduced overall root mean squared difference (RMSD) from 5.61 ± 4.11° to 3.12 ± 1.88° and significantly reduced systematic differences. For example, ankle rotation RMSD during walking decreased from 12.71° to 4.94° in males, and from 16.14° to 6.11° in females. Despite improvements, females generally exhibited larger RMSD between ML and MB at several joints, including pelvis rotation, ankle inversion and ankle rotation, suggesting a sex-specific effect in the comparison of MB and ML systems. Although LCS alignment substantially improved agreement, residual discrepancies remained.

Overall, both MB and ML systems identified key sex-based gait and landing characteristics. Both systems consistently identified greater pelvic list, hip adduction, and ankle flexion in females during walking, as well as greater anterior pelvic tilt, knee abduction, and ankle flexion in females during the drop vertical jump. However, agreement was generally higher in the sagittal plane, whereas lower agreement in non-sagittal plane kinematics - especially among females - highlights the need for careful consideration when choosing between ML and MB motion capture for research or clinical applications that rely on detecting subtle non-sagittal plane mechanics.