Geometry-informed correction of projection bias in browser-based monocular squat assessment.

Introduction: Monocular RGB-based human pose estimation is increasingly applied in field-based movement assessments; however, systematic projection-related errors inherent to the simplified camera geometry remain insufficiently characterized relative to laboratory-based biomechanical standards. This study quantified systematic projection bias in a browser-based markerless motion capture (MMC) system and validated a geometry-informed linear correction framework for sagittal-plane squat analysis under standardized monocular acquisition conditions. Methods: Bodyweight squats performed by 30 healthy adult males were simultaneously recorded using a three-dimensional marker-based optical motion capture (OMC) system and a two-dimensional webcam positioned at a fixed distance and orientation. Results: Raw joint angles obtained from the MMC system revealed significant systematic underestimation (hip: -11.2°, knee: -10.6°), consistent with perspective-induced projection effects predicted by the pinhole camera model and differences in the joint center definitions. To address this projection-consistent bias, a geometry-constrained linear correction model was developed and validated using leave-one-subject-out cross-validation. The correction effectively neutralized the systematic bias (hip: 0°, Discussion: Importantly, high accuracy was achieved without incorporating participant-specific anthropometric variables, suggesting that projection-consistent geometric factors predominated under controlled camera conditions. These findings demonstrate that systematic errors in monocular pose estimation can be substantially mitigated when the acquisition distance and orientation are standardized. Our results suggest that under such predefined recording constraints, corrected joint angles provide practically relevant estimates for strength assessment and rehabilitation monitoring.