Figure 12.

Cycle-based data (medians) from the complete trials of all subjects. (Top) The estimated upper arm heading offset ($\widehat{\delta }$) agrees well with the true offset δ between the upper arm headings of the optical reference and of the uncorrected orientation estimate by the IMU (all as medians of each trial iteration). (Bottom) Errors of the shoulder position with respect to the opt. ref. as medians of each trial iteration. In the clear majority (91 %) of all trials of all subjects, the median error of the proposed method is smaller than five centimeters, while the conventional method (Conv.) fails to determine adequate shoulder positions during comp. mov. and in the transition phase. (Right) Corresponding violin plots of the plots on the left excluding the transition phase. Upper/lower whiskers of the inner box plots show the 95th/5th percentiles. Horizontal red lines indicate median values over all subjects.