Fig 1.
Illustrating: A) abduction and adduction, B) metacarpophalangeal flexion, and C) thumb opposition.
Fig 2.
Video recordings captured from frontal view.
Twelve healthy participants at the Upper Limb Motion Analysis Laboratory at Imperial College London.
Fig 3.
Output from OpenPose that illustrates the inferred keypoints overlapped onto the image frames for four representative participants.
Fig 4.
Butterworth filter in output from OpenPose.
Butterworth filter with 1 Hz, 2 Hz and 3 Hz cut off frequencies (c/o freq.) applied to the OpenPose signal of the thumb interphalangeal joint angle for one representative participant.
Fig 5.
Abduction and adduction angles.
Measured position for the finger intersect joint of the index finger (A), and of the thumb (B).
Fig 6.
Measured position for the metacarpophalangeal (MCP) joint of the index finger (A), and of the thumb (B). Measured angles of the proximal interphalangeal (PIP) joint of the index finger (C), and of the thumb (D).
Fig 7.
Geometric representation of the finger.
Illustration of a geometric representation of the finger, where WRST indicates the wrist, MCP indicates the metacarpophalangeal joint, PIP indicates the proximal interphalangeal joint, and DIP indicates the distal interphalangeal. α represents the included angle of the metacarpophalangeal joint, β represents the included angle of the proximal interphalangeal joints, and γ represents the included angle of the distal interphalangeal joints.
Fig 8.
Examples of raw data for one healthy participant.
Examples of averaged raw data for (A) 2nd-to-3rd digit angle for four repetitions of the abduction and adduction task, (B) 2nd-to-3rd finger angle for two repetitions of the radial walking task, (C) 2nd metacarpophalangeal (MCP) joint angle for two repetitions of the MCP flexion task, (D) 2nd proximal interphalangeal joint angle for the thumb opposition task, estimated using OpenPose (ML; solid lines) and measured with the optical motion capture system (QTM; dashed lines) for one representative healthy participant.
Fig 9.
RMSEs for the four activities.
Root mean square error (RMSE) differences between OpenPose on monocular images and marker-based optical motion capture system during A) finger abduction and adduction, B) radial walking, C) finger metacarpophalangeal flexion, and D) thumb opposition. Each colour represents a different participant.
Fig 10.
BA plots showing the total active flexion mean for A) abduction and adduction, B) radial walking, C) metacarpophalangeal flexion, and D) thumb opposition of the 2nd, 3rd, 4th and 5th digits for the metacarpophalangeal and proximal interphalangeal joints of the left and the right hands.
Fig 11.
Linear regression plots of total active flexion for (A) abduction and adduction, (B) radial walking, (C) MCP flexion and (D) thumb opposition of the 2nd, 3rd, 4th and 5th digits for the metacarpophalangeal and proximal interphalangeal joints of the left and the right hands.