Methods of measurement
The CV-aided system approach (Good Boost CV system, Good Boost Wellness,
UK, 2021) in this study involved a modified version of OpenPose, a
computer vision algorithm trained to detect key landmarks on the human
body within camera images. For a given frame of image/video data,
OpenPose returns predicted x,y coordinates for each body part and each
human detected in the image. X,y coordinates were used to compute
metrics such as joint angles and distances (in pixels) between two body
parts for the index of movements. To translate distance values into
real-world distances, at the start of each movement, the participant or
investigator held up a calibration checkerboard parallel to the camera
and at the same distance at which the movement was performed; Python’s
OpenCV package was used to automatically detect the corners of the
checkerboard to scale all distance values from pixels to centimetres.
The videos taken in the movement laboratory were captured by a Logitech
C920 pro HD webcam (©2021 Logitech, UK) with 1080p resolution and 30
frames per second sampling rate. The videos taken in the home setting
were captured by the participant’s smartphone camera, tablet camera or
webcam. Spinal curvature was measured in the laboratory only using a
portable surface topography method employing the Microsoft Kinect sensor
V2 (Microsoft Corporation, Seattle, Washington, U.S.A) and using an
established method to measure thoracic kyphosis. The reference tests
were a series of standard clinical assessments measured by an
experienced physiotherapist who was blinded to the remote technology
systems analyses and results.