<![CDATA[Computer-vision-based markerless pose estimation is increasingly used to quantify sport movement outside laboratory settings, but its validity is not uniform across tasks, joints, anatomical planes, or environmental conditions. This systematic review synthesised peer-reviewed evidence published from 2020 to 2025 using PRISMA 2020 procedures and a PICO-based search across Scopus, PubMed, and ScienceDirect. Eighteen studies met the final eligibility criteria, including 17 empirical studies and one narrative review. A narrative thematic synthesis was combined with methodological appraisal and an operational validity framework linking target variables, reference standards, and intended use. Across validation studies, lower-limb sagittal-plane agreement reached CMC > .90 in the Pose2Sim study, whereas real-competition OpenPose validation reported CMD about .73 for knee-angle waveforms and weaker agreement for hip and ankle measures. A review-level estimate included in the corpus placed markerless sagittal-plane error at approximately 3–15°, with larger errors in the transverse plane. The synthesis therefore supports markerless systems most strongly for spatiotemporal, gross-displacement, and selected sagittal-plane variables when task-specific validation demonstrates small systematic bias. Transverse-plane, rotational, occlusion-prone, and high-speed measures remain conditional and require additional validation. Practitioners should select systems according to the target variable and intended decision, rather than treating markerless capture as a single accuracy class. Future research should standardise validation metrics, report absolute error and bias alongside agreement statistics, and test systems under genuine competition conditions.]]>
Copyrights © 2026
