Video-Based Markerless Motion Capture for Clinical and Rehabilitation Biomechanics: A PRISMA-ScR Scoping Review of Validated Architectures, Clinical Readiness, and Emerging Methods

Video-based markerless motion capture (MMC) promises to remove the cost, space, and skin-marker constraints of traditional optoelectronic systems. However, it remains unclear whether current pipelines provide clinically acceptable biomechanical data. This scoping review, conducted according to PRISMA-ScR guidelines, mapped 117 validated biomechanical studies (the MACRO corpus) against a curated, non-exhaustive scan of emerging computer-vision research (the MICRO corpus) to identify the translational gap between technical innovation and clinical readiness.

The evidence base is currently narrow, dominated by healthy adults walking in laboratory settings. Pipelines are categorized into five architectural families, yet most stop at raw geometric joint angles without biomechanical refinement. Metrological validation shows that sagittal lower-limb joint-angle errors typically cluster around 5 to 6 degrees, which exceeds the threshold for clinical reliability. Crucially, out-of-plane kinematics (frontal and transverse planes) and kinetics—which are vital for diagnosing neurological gait disorders—are systematically under-reported or unvalidated. While computer-vision research has advanced rapidly with foundation-model mesh recovery, differentiable inverse kinematics, and video-based kinetics, these building blocks are almost entirely absent from the validated biomedical literature.

For clinicians, spatiotemporal parameters (such as walking speed) are currently the only outputs ready for routine clinical use. Sagittal joint angles may be useful for tracking gross qualitative trends but are not yet reliable for detecting small, clinically significant changes. Frontal and transverse plane kinematics and kinetics remain strictly research-grade. The review highlights that deployability and validation-readiness are currently inversely related: the most accessible monocular tools are the least refined, creating a risk that systems may be deployed in clinics before they are validated for the specific decisions they are intended to support.