Fig 1.
Overview of laboratory space and workflow.
A) Representative image frame of original video recording from left-side sagittal plane view with diagram of motion capture and video cameras. We used a known distance of 6.30 m between two strips of tape on the floor to dimensionalize pixel coordinates of OpenPose keypoints. The public dataset [18] that we used is made available at http://bytom.pja.edu.pl/projekty/hm-gpjatk/. See Release Agreement for copyrights and permissions. B) Workflow of video recordings (available at https://github.com/janstenum/GaitAnalysis-PoseEstimation). We analyzed video recordings with OpenPose using Google Colaboratory and next processed the data using custom MATLAB scripts.
Table 1.
Differences in event times for all steps.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view. Asterisks (*) denote P < 0.05.
Fig 2.
Temporal gait parameters for all individual steps shown for all participants and measurement systems.
Panels show A) step time, B) stance time, C) swing time, and D) double support time. Data shown in blue represent comparisons between motion capture and OpenPose left-side (CL) views, data shown in red represent comparisons between motion capture and OpenPose right-side (CR) views, and data shown in gray represent comparisons between the two OpenPose views (CL and CR). Dark circles represent left leg data, light circles represent right leg data. Color schemes and shading are consistent across all similar figures (Figs 3 and 4). Bar plots on the far right show individual data, group means, and SD to visualize the distribution of the differences observed between the measurement systems. Results from statistical analyses are shown in Tables 2 and 3.
Table 2.
Gait parameters calculated for all steps.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view.
Table 3.
Statistical tests for gait parameters calculated for all steps.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view. Asterisks (*) denote P < 0.05.
Fig 3.
Temporal gait parameters calculated as individual participant means shown for all participants and measurement systems.
Panels show A) step time, B) stance time, C) swing time, and D) double support time. Bar plots on the far right show individual data, group means, and SD to visualize the distribution of the mean differences observed between the measurement systems. Results from statistical analyses are shown in Tables 4 and 5.
Table 4.
Gait parameters calculated as individual participant means.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view.
Table 5.
Statistical tests for gait parameters calculated as individual participant means.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view. Asterisks (*) denote P < 0.05.
Fig 4.
Step length and gait speed comparisons among the different measurement systems.
A) Step lengths calculated for all individual steps for all participants and measurement systems. B) Step lengths calculated as individual participant means for all participants and measurement systems. C) Gait speeds calculated as individual participant means for all participants and measurement systems. Bar plots on the far right show individual data, group means, and SD to visualize the distributions of the differences observed between the measurement systems. Results from statistical analyses are shown in Tables 2–5.
Fig 5.
Differences in step lengths calculated from each measurement system in relation to anterior-posterior position on the walkway.
Dark circles represent left step lengths, light circles represent right step lengths. A) Differences in step lengths calculated by motion capture and OpenPose left (CL) side views across the walkway. B) Differences in step lengths calculated by motion capture and OpenPose right (CR) side views across the walkway. C) Differences in step lengths calculated by OpenPose left (CL) and right (CR) side views across the walkway.
Fig 6.
Example image frames of step length errors.
Representative image frames taken from OpenPose output videos highlighting the discrepancies in step length calculation at different positions along the walkway and from different camera views. Note that images from camera CR are reflected across a vertical axis to make the walking direction consistent across camera views. The public dataset [18] from which the images belong is made available at http://bytom.pja.edu.pl/projekty/hm-gpjatk/. See Release Agreement for copyrights and permissions.
Fig 7.
Sagittal plane hip, knee and ankle angles.
A) Group mean ± SD ensemble sagittal left and right joint angles for the three measurement systems: motion capture (MC), OpenPose left-side view (CL) and right-side view (CR). For all angles, positive values indicate flexion (or dorsiflexion) and negative values indicate extension (or plantarflexion). B) Cross-correlation coefficients at time lag zero (individual data, group means and SD) for individual participant mean joint angle profiles between measurement systems.
Table 6.
Mean absolute error comparisons of joint angles.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view.
Table 7.
Cross-correlations of joint angles.
MC: motion capture; CL: OpenPose left-side view; CR: OpenPose right-side view.
Fig 8.
Effect of walkway position on sagittal plane joint angles.
Comparisons of joint angles and mean absolute error (MAE) between measurement systems at three portions of the walkway. The portions were based on the average anterior-posterior position of the C7 marker in motion capture data throughout each gait cycle. Gait cycles were binned in regions: ‘Start’ spanning anterior-posterior positions less than −0.50 m of the middle of the walkway, ‘Middle spanning position from −0.50 to 0.50 m, and ‘End’ spanning positions greater than 0.50 m. The number of gait cycles analyzed were 30, 22 and 24 for ‘Start’, ‘Middle’ and ‘End’, respectively, for the left leg and 31, 21 and 26 for the right leg.