Fig 1.
Two-dimensional landmarks for BlazePose (left) and the marker set used for Opti- Track’s optical motion capture system (right). BlazePose produces 33 landmarks, whereas OptiTrack’s marker set contains 39 markers. Markers relevant to the range of motion calculation are marked with their respective names. Figure in the middle is a screenshot of the landmarks detected by MediaPipe’s BlazePose overlayed on a video image captured during data collection with the participant wearing a motion capture suit.
Table 1.
Tested joint range of motion movements.
Fig 2.
An illustration of the seasonality decomposition to identify anomalies in the movement angle data. Top panel: The original data contain three noticeable outliers that may affect the subsequent local maxima detection. Middle panel: The derived trend from the seasonality decomposition. Note that the decomposition could potentially introduce artifacts that alter the values of the angle, which was why it was not directly used to identify the maximum angle. Bottom panel: The residuals from the seasonality decomposition where the locations of the three outliers are apparent.
Table 2.
Intra- and inter-rater reliability measures.
Fig 3.
Regressions between the range of motion (ROM) angles derived from OptiTrack (x-axis) and BlazePose (y-axis) across all participants for each joint. Different markers represent different joint movements. The dotted lines in each subplot are reference lines and have a slope of 1 and an intercept of 0.
Fig 4.
Sample movement angles of elbow flexion.
Sample movement angles as a function of time for elbow flexion from OptiTrack (top row) and BlazePose (bottom row).
Fig 5.
An experimenter performing the elbow flexion movement at the starting (left) and maximally flexed (right) positions with an overlay of BlazePose’s estimated joint locations.