100 Section - Validation of markerless motion capture system for temporal and spatial parameters during the Timed Up and Go test

Objectives : The Timed Up and Go test (TUG) is widely used to assess functional mobility, but total time alone provides limited information on movement quality. Markerless motion capture (MLS) is a practical alternative to marker-based systems; however, the agreement and error of temporal and spatial TUG parameters remain unclear. This study evaluated agreement and error characteristics of selected TUG parameters using an AlphaPose-based MLS system with a Vicon system as reference.

Design: Twenty-six healthy adults performed three TUG trials at a self-selected pace. MLS data were acquired using three iPhones (60 Hz), and two-dimensional joint landmarks estimated with AlphaPose were reconstructed into three-dimensional coordinates via triangulation. Simultaneously, marker-based motion capture data were collected using a 10-camera Vicon system (100 Hz). Outcome measures were total TUG time, center-of-mass velocity, stance time, step length, and step width. Agreement and measurement error were evaluated using Bland–Altman analyses and limits of agreement, and consistency was assessed using the intraclass correlation coefficient [ICC(3,1)].

Results: Total TUG duration showed a small mean bias and moderate agreement (ICC = 0.69). Center-of-mass velocity showed a negative bias with excellent agreement (ICC = 0.87). Stance time demonstrated minimal bias and relatively narrow limits of agreement but low consistency (ICC = 0.16). In contrast, spatial parameters showed larger absolute errors, with wide limits of agreement for both step length and step width.

Conclusion:

An AlphaPose-based MLS system demonstrated acceptable performance for selected temporal TUG outcomes, particularly total TUG time and center-of-mass velocity, supporting their use for group-level analyses. In contrast, spatial gait parameters showed substantial absolute error and wide limits of agreement, limiting their suitability for individual-level quantification. These findings indicate that temporal parameters are more robust than spatial measures in open-source MLS-based TUG analysis, and that group- and individual-level interpretations should be clearly distinguished.