Date of Award

2024

Degree Name

Biomechanics

College

College of Health Professions

Type of Degree

M.S.

Document Type

Thesis

First Advisor

Dr. Steven Leigh, Committee Chairperson

Second Advisor

Dr. Suzanne Konz

Third Advisor

Mr. Brian Morgan

Abstract

While 3D direct linear transformation is a versatile motion capture method, care must be taken so that the cameras are not bumped or adjusted between the object space calibration phase and the end of video recording. Disturbing the cameras during this time will misalign the view, making the movement data invalid for the associated calibration. The purpose of this study was to design and validate a cross-platform software application for transforming image coordinates from a bumped camera to be compatible with the object space calibration and to determine the relationship between reconstruction error and changes to bumped camera views.

For the validation test, a cubic object space was defined by four survey poles. A reference line used to guide the transformations was represented by two markers inside of the object space. A physical reference object of known, fixed dimensions was placed inside the object space. The positions, orientations, and zoom amounts of one of two cameras facing the object space were changed by discrete amounts of up to 10 cm translation, 10° rotation, and ±50% magnification before recording to simulate bumps of the camera and an unwanted change in the camera view. Both cameras were then moved to the most extreme positions defined by the maximum translation, rotation, and magnification before recording. A frame from each video of every trial was extracted, digitized, and processed to obtain landmark coordinates of the survey poles and reference object. The frames representing the cameras’ neutral positions were designated as calibration frames. The application transformed the image coordinates for the bumped trials so that the trial and calibration reference lines overlapped. Errors between the known and calculated physical dimensions of the survey poles and reference object were calculated to determine the relationships between error and combinations of camera position, orientation, and magnification.

Error values were below the minimum detectable change of 36 mm. Object axis errors were reduced to about 1.1% when the application corrected trials in which the camera was perturbed. No practically significant error patterns or changes in the shape of errors across camera translation, rotation, or magnification were found. Global Y was consistently higher than global Z, which was consistently higher than global X. This matched the pattern of pole deviations in which global X was collinear with both poles that defined it, global Z deviated by 1°, and global Y deviated by 2°. The axes of the camera and the object space were not aligned, which, when coupled with the discretization of the projected lengths caused by finite resolution, resulted in a greater range of error in object x and for all axes when zoomed out.

All hypotheses were supported; error magnitudes were reduced without meaningful patterns. The application transformations reduced 2D and 3D coordinate error enough for the trials to be suitable for analysis. This means that in future biomechanics studies, fewer trials in which camera bump occurred will be discarded, resulting in more data available from which to draw conclusions.

Subject(s)

Three-dimensional imaging in medicine.

Photogrammetry.

Biomechanics.

Kinesiology.

Physics and applied physics.

Download

Share

COinS