Table 1.
Terrain type segments for course A.
Table 2.
Number of steps by walk mode.
Fig 1.
Map of walking course A, including markings and labeling of terrain type segments.
Fig 2.
Eye pitch angles were calculated by linear interpolation form the normalized (x, y)-positions of the gaze points in the world camera image frames that are available from the eye tracker. Positive eye pitch angles correspond to looking up. Negative eye pitch angles correspond to looking down.
Fig 3.
(A) A positive head pitch angle corresponds to looking up. (B) A neural head pitch angle corresponds to looking straight ahead. (C) A negative head pitch angle corresponds to looking down.
Table 3.
Motion and gaze variables used in the analysis.
Table 4.
Overview over the walk mode transitions from the multi-modal gait database [4] that were analysed in this study.
Fig 4.
Illustration of the classification task.
The rows in the sub-figures represent the data of a transition. Blue squares represent steps before a walk mode transition, gray squares represent steps after a walk mode transition. The sub-figures show samples obtained by sliding a moving window of size two (red frames) from left to right towards the walk mode transition that is indicated by a change in color from blue to gray squares. The corresponding class label is defined by the blue framed squares: it is one if a walk mode transition occurs in (A) the next step, (B) two steps, (C) three steps, and (D) in 4 steps. Otherwise the label is zero.
Fig 5.
Gaze and gait parameters for stairs transitions as a function of steps to/from walk mode transition.
Left row: Deviation of eye and head pitch angles from the baseline for transitions involving stairs. Right row: Deviations of the normalized step length and period from their baselines for transitions involving stairs. The error bars indicate the 95 % confidence interval of the non-parametric bootstrap distribution. Stars indicate statistically significant differences with a significance level . The transition is marked by a dotted line. The shaded area between 1 step before and after the transition is subject to ambiguities in the walk mode labeling.
Fig 6.
Gaze and gait parameters for ramp transitions as a function of steps to/from walk mode transition.
Left row: Deviation of eye and head pitch angles from the baseline for transitions involving ramps. Right row: Deviation of eye and head pitch angles from their baselines for transitions involving ramps. The error bars indicate the 95 % confidence interval of the non-parametric bootstrap distribution. Stars indicate statistically significant differences with a significance level . The shaded area between 1 step before and after the transition is subject to ambiguities in the walk mode labeling.
Fig 7.
Forecasting results for transitions to walk.
The left column shows the forecasting results of transitions to walk in terms of Matthews’ correlation coefficient for various forecasting horizons and different input parameters for transitions involving stairs. The right column shows the same information for transitions involving ramps. Error bars indicate bootstrapped 95 % confidence intervals. Stars indicate statistically significant differences with a significance level .