Fig 1.
Types of anomalies during gait using OpenPose.
The left panel shows the OpenPose skeletal model and the name of each part. The right panel shows the anomaly types corresponding to those in Table 1. Estimation accuracy in Table 1 is excluded from the figure for convenience of illustration. ROM, range of motion; COG, center of gravity.
Table 1.
Percentage of each anomaly type.
Fig 2.
Histograms of ROM of hip and knee joints (left direction: red, right direction: blue).
The ranges in red and blue shading are 95% confidence intervals. The black shaded area indicates the maximum and minimum ROM during gait using inertial sensors as reported by Park et al. ROM, range of motion [20].
Table 2.
Mean shift and variability of ROM at the hip and knee joints, comparing published gyro-sensor-based statistics for ROM with those obtained in the present database analysis.
Fig 3.
(a) Distribution of reliability scores for all subjects. Two groups, low confidence (blue) and high confidence (red), were assumed and classified by clustering. (b) Reliability score per part. The upper panel shows the left direction, and the lower panel shows the right direction.
Fig 4.
Estimated accuracy for each part with OpenPose.
The accuracy based on the anomaly types (i.e., the percentage not containing any of the anomalies listed in Table 1) is shown. The numbers in parentheses represent the accuracy after correcting the proposed workflow.
Fig 5.
Proposed workflow.
Fig 6.
The left and right panels show the skeletal lengths of the thigh and lower leg, respectively, which depend on the distance between the camera (from the right side) and the subjects. The red and blue represent the right and left sides of the body, respectively.
Table 3.
Sensitivity of detection of each type of anomaly by workflow.
Table 4.
Reproducibility of values by workflow.