From linear to nonlinear gait measures: validity of a multi-view markerless motion capture system

Markerless motion capture offers a practical alternative to marker-based optoelectronic systems, yet validation studies have focused almost exclusively on linear gait measures. Nonlinear measures of gait dynamics, sensitive to the fine temporal structure of locomotor signals, remain unvalidated in markerless systems. This study assessed the concurrent validity of a three-camera markerless system against a 15-camera optoelectronic system across four treadmill speeds in 22 healthy adults. Inter-system agreement was evaluated for spatiotemporal parameters (mean, variability, detrended fluctuation analysis [DFA] scaling exponents) and for joint angle and trunk acceleration time series (maximum Lyapunov exponents, sample entropy, Attractor Complexity Index [ACI]). Temporal measures reached near-perfect agreement, and spatial measures showed good to excellent agreement with small speed-dependent positive biases. Sagittal-plane joint kinematic waveforms were compared using statistical parametric mapping, with root-mean-square error (RMSE) reported for significant intervals. Agreement was best at the hip, with knee and ankle showing comparable, higher error (RMSE: 1.2-3.0{degrees} hip, 3.7-5.5{degrees} knee and 3.8-5.1{degrees} ankle). ACI demonstrated moderate to good agreement across all joints and good agreement across trunk acceleration directions. Most DFA scaling exponents for step-based series supported group-level comparisons: both measures are usable for markerless assessment of gait's nonlinear dynamics. Maximum Lyapunov exponents and sample entropy showed lower absolute agreement: at the hip and knee, they preserved inter-individual ranking and remained usable for within-system group comparisons, but agreement collapsed at the ankle and for trunk sample entropy, indicating these measures still need refinement. These findings define a tiered, measure-specific scope for markerless gait analysis, extending validation beyond spatiotemporal parameters.