Assistant Professor Sacred Heart University Fairfield, Connecticut, United States
Abstract Body : In this study, we evaluate the reliability of the Theia Markerless motion capture system for determining anthropometric measurements in a group of 15 healthy adults. Linear anthropometric measures of each of subject’s body and limb segments were first taken manually using an anthropometer (torso height, shoulder & pelvic breadth, arm, forearm, hand, thigh, shank, foot lengths). A 10 second video of the subject standing still was then recorded using eight cameras positioned for a 360-degree perspective, and segment length measurements were calculated from the segment endpoints as determined by the Theia Markerless software.
Pearson correlations and two-way intraclass correlation coefficients (ICC3) were calculated between the manual and Theia measurement sets in their entirety and by segment. When the entire data set was considered, the measures calculated from the Theia data demonstrated a high positive Pearson correlation (0.948) and high reliability (ICC3 = 0.977, CI= 0.93, 0.99) with hand-measured anthropometric measurements. The correlations dropped when considered at the by-segment level, with the lowest correlations obtained for the right & left thigh length (0.402) and the highest correlations obtained for the right- and left-hand lengths (right = 0.846, left = 0.875). A similar pattern emerged in the by-segment reliability scores (ICC3 = 0.294,0.402 for thighs, ICC3 = 0.821, 0.821 for hands). The difference between across and by- segment correlations may be due to insufficient statistical power (power = 0.97 across segments, and a range of powers from 0.204 to 0.991 by-segment) with our sample size of 15.
Moving forward, we plan to expand our database to increase the power of our by-segment ICC3 results and produce a clearer assessment about whether markerless motion capture can reliably supplant manual measurements for anthropometry.