Fig 1.
Movement patterns and gap distributions in CP-gait and WalkRun.
Top row: Illustration of the movement patterns. Note that certain maker trajectories have been merged to reduce the number of markers. Bottom row: Examples of the measurement matrices of the two movement patterns after being corrupted with gaps. Left matrix: CP-gait, right matrix: WalkRun.
Table 1.
Mean Euclidean differences between reconstruction and measurement for gaps in single markers.
Fig 2.
Reconstruction of gaps in single markers in the CP-gait dataset.
Shaded red area indicates the gapped time frames. Blue lines represent reconstructed data, black lines input data, dotted black lines reference trajectory. The gridline spacing in the callout-figures is 1 cm.
Fig 3.
Reconstruction of gaps in single markers in the WalkRun dataset.
Shaded red area indicates the gapped time frames. Blue lines represent reconstructed data, black lines input data, dotted black lines reference trajectory. The gridline spacing in the callout-figures is 1 cm.
Table 2.
Mean differences between reconstruction and measurement for gaps in single markers in the WalkL dataset.
Fig 4.
Boxplots of reconstruction accuracies.
R1 and R2 are the two algorithms proposed in the current study. Spline interpolation and the DynaMMo algorithm [7] are included for comparison.
Fig 5.
Example reconstruction of a gap in the left tibia marker in the WalkL dataset.
The black lines represent the reference trajectories, the blue lines represent reconstructed trajectories for each method. Each reconstruction was shifted by 15 mm vertically for better clarity.
Fig 6.
Results of sensitivity analysis.
wmm: weight factor on missing marker coordinates, σ: scaling of weight factors, ϴλ: threshold on cumulative sum of singular values, ϴD: threshold to discard distal corrupted trajectories. Error bars show the standard error of the mean.