
Pose QA
PoseQA applies a series of heuristics to determine whether a person is detected with confidence and is fully in frame for most of the movement. This process ensures that keypoint quality is sufficient to continue with movement analysis. Broadly, these checks involve:- having sufficient overlap of frames where a person is detected in both camera views
- a person is detected in-frame for a certain percentage of the movement duration
- keypoint confidence values meet minimim thresholds
- ensuring projection of 3D keypoints and 2D detection/keypoints align with expected distributions for each body segment
Biomech QA
Biomech QA tests the calculated values of events and metrics for a specific movement-activity. Note:other-other
(general movement) does not have any Biomech QA tests.
For example, baseball-pitching
biomech QA looks for:
- Events
- certain events present: [twist, max knee raise, foot contact, release]
- events occur within a 2 s window: [max knee raise, foot contact, release]
- Metrics
- pelvis and trunk angular velocities are not excessively noisy (via rolling standard deviation tolerance)
- x_factor near foot contact event not greater than 120 deg
- forearm length (dominant arm distance between elbow and wrist joint centers) not greater than 0.5 m
- peak_pelvis_angular_velocity between 200 and 1000 deg/s
- peak_trunk_angular_velocity between 300 and 1500 deg/s
- peak_arm_angular_velocity between 500 and 2500 deg/s
jump-countermovement
Biomech QA looks for:
- Events
- all events are present: [initiation, bottom, takeoff, peak jump, landing, landing bottom, termination]
- ensure events occur in the correct order: [initiation -> bottom -> takeoff -> peak jump -> landing -> landing bottom -> termination]
- Metrics
- check that metrics have reasonable values:
- jump height between 0.075 m (~3 inches) and 1.3 m (~51 inches)
- contraction time (from initiation to takeoff) less than 2.5 seconds (otherwise its not a countermovement jump)
- hip knee dominance (for eccentric, concentric, and landing phases) between 0 and 1
- bilateral thigh length standard deviations under 0.065 m (~2.5 in)
- check that metrics have reasonable values: