Automatic Detection Of Running Gait Events From Marker-less Motion Capture Data

Accurate determination of gait events in the running cycle (foot strike, FS; toe off, TO) is critical for biomechanical analysis. In the absence of a force measurement device, kinematic data from marker-based motion capture (MOCAP) has been used to identify these events. However, no investigation has explored the feasibility of automatic event detection from a marker-less (ML) MOCAP system. PURPOSE: To determine the level of agreement between spatiotemporal metrics (cadence, stance time, step length) derived from automated detection of gait events using ML MOCAP and a pressure-sensitive treadmill (PT). METHODS: Nineteen experienced runners (13 F, 19.7 ± 1.4 yrs) performed three consecutive two minute running trials on a PT (100 Hz) at self-selected speeds corresponding to ratings of perceived effort of 3, 5 and 7-out-of-10. Runners used a self-selected foot strike (11 rearfoot, 8 forefoot) and wore their own shoes. Pressure data was collected for the last 25 seconds of each trial and eight synchronized cameras (120 Hz) recorded video data. Video data was processed with commercial software to determine 19-segment 3D model pose estimates and whole-body center of mass (COM) trajectory throughout each trial. Vertical COM velocity and the vertical trajectory of the toe segment’s distal end point were utilized to determine FS and TO timing, respectively. Cadence, stance time and step length were computed and compared to metrics computed from PT. Comparisons were made using mean difference (MD), Bland-Altman limits of agreement (LOA) over 95% confidence interval, and interclass correlation coefficients (ICCs). RESULTS: Mean cadence (ML: 170.2 ± 9.3 steps/min, PT: 170.0 ± 9.3 steps/min; MD = -0.2; LOA: [-0.79, 0.49]; ICC = 1.0), mean stance time (ML: 0.236 ± 0.020 s, PT: 0.234 ± 0.022 s; MD = -0.001; LOA: [-0.014, 0.012]; ICC =0.982), and mean step length (ML: 1.21 ± 0.12 m, PT: 1.21 ± 0.12 m; MD = 0.05 cm; LOA: [-0.74, 0.83]; ICC = 1.0) demonstrated excellent agreement across all trials (n = 57). CONCLUSION: In a population of distance runners with heterogenous foot strike, non-standardized footwear and across a range of speeds (2.67 - 4.44 m/s), a marker-less motion capture system accurately assessed spatiotemporal metrics when compared to a pressure-sensitive treadmill, suggesting that automatic gait event detection is accurate.