Fig 1.
Experimental setup for (A) forward and backward and (B) sideways falls illustrating Kinovea traces from digitization of key anatomical locations. Right panels display digitized traces from Kinovea of the time-varying positions during the falls of the head (pink), shoulder (green), elbow (purple), wrist (yellow), hip (light blue), knee (red), ankle (grey) and sternum (dark blue). Pink tape of the mat displays the plane of the fall. We examined how the accuracy of kinematic outcomes from Kinovea depended on (C) calibration technique (2D calibration grid versus 1D calibration based on participant height), (D) calibration frame translation (purple grids for forward falls and green grids for backward falls), (E) calibration grid rotation (blue grids), and camera view with respect to the fall (30, 60 and 90 degree views shown in (A), (C), and (B), respectively).
Fig 2.
Time-varying linear positions and velocities measured from 3D motion capture (Qualisys) and Kinovea.
The images show perturbation-based falls experienced by a 32 year old woman in the (A) backward fall direction, (B) forward fall direction, and (C) sideways fall direction. All traces are between onset of platform movement (time = 0) and end of fall (cessation of movement). Raw unfiltered data show considerable non-physiological noise. A 20 Hz low pass filter of Qualisys position data retained 96% of the energy content for 96.7% of the vertical velocity traces.
Table 1.
Errors between position and velocity signals from Qualisys and Kinovea from different cut-off frequencies.
Fig 3.
Angular positions and velocities measured from 3D motion capture (Qualisys) and Kinovea.
The traces show time-varying angular positions (with respect to the horizontal) and velocities for selected body segments, measured for (A) backward, (B) forward, and (C) sideways perturbation-based falls. Data are for the same falls as shown in Fig 2.
Fig 4.
Effect of low-pass filter cut-off frequency on the accuracy of Kinovea estimates.
(A) Linear position, (B) linear velocity, (C) angular position, and (D) angular velocity. The plots show root mean square errors (RMSE) for Kinovea data filtered at cut-off frequencies of 3, 5, 7, 10, 12 and 14 Hz, based on comparison with Qualisys ground truth signals. Results are specific to Kinovea data from the 90 deg camera angle, calibrated with the calibration frame located in the plane of the fall. Data points show mean values averaged across all body parts and fall directions with error bars showing ± 1 SE. * p≤0.05, ** p≤0.01, *** p≤0.001.
Fig 5.
Effect of fall direction on accuracy of Kinovea estimates.
(A) Linear position, (B) linear velocity, (C) angular position, and (D) angular velocity. RMSEs are based on comparison of Qualisys ground truth signal to Kinovea data from the 90 deg camera angle, calibrated with the calibration frame located in the plane of the fall, with position data filtered at 10 Hz. Data points show mean values averaged across all body parts with error bars showing ± 1 SE. * p≤0.05, ** p≤0.01, *** p≤0.001.
Fig 6.
Effects of camera angle and fall direction on agreement between Kinovea and Qualisys peak velocities.
The graphs compare peak vertical and horizontal velocities for various anatomical locations. Results are specific to the calibration grid located in the plane of the fall and Kinovea velocities differentiated from position data filtered with a cut-off frequency of 10 Hz.
Fig 7.
Differences across body sites and fall directions in the accuracy of Kinovea estimates.
(A) Linear position, (B) linear velocity, (C) angular position, and (D) angular velocity. Root mean square errors (RMSE) are from comparison between Qualisys ground truth signals and Kinovea estimates based on the 90 deg camera angle, the calibration frame located in the plane of the fall, and filtering of position data with a 10 Hz cut-off frequency. Data points show mean values, and error bars show ± 1 SE (sideways = blue, backward = purple, forward = orange; grey = mean values across all fall directions). Significant differences are shown for mean values across all fall directions (* p≤0.05, ** p≤0.01, *** p≤0.001).
Fig 8.
Effect of camera angle on the accuracy of Kinovea estimates.
(A) Linear position, (B) linear velocity, (C) angular position, and (D) angular velocity. Root mean square errors (RMSE) are from comparison between Qualisys ground truth signals and Kinovea estimates based on the calibration frame located in the plane of the fall, and filtering of position data with a 10 Hz cut-off frequency. Data points show mean values, and error bars show ± 1 SE. * p≤0.05, ** p≤0.01, *** p≤0.001.
Fig 9.
Effect of calibration technique on the accuracy of Kinovea estimates.
(A) Vertical velocity, (B) horizontal velocity, and (C) angular velocity. Root mean square errors (RMSE) are from comparison between Qualisys ground truth signals and Kinovea estimates from the 90 deg camera angle, and filtering of position data with a 10 Hz cut-off frequency. Data points show mean values, and error bars show ± 1 SE. Statistical comparisons in two left columns are based on mean values. * p≤0.05, ** p≤0.01, *** p≤0.001.