Visual Demands of Walking Are Reflected in Eye-Blink-Evoked EEG-Activity

Abstract

Blinking is a natural user-induced response which paces visual information processing. This study investigates whether blinks are viable for segmenting continuous electroencephalography (EEG) activity, for inferring cognitive demands in ecologically valid work environments. We report the blink-related EEG measures of participants who performed auditory tasks either standing, walking on grass, or whilst completing an obstacle course. Blink-related EEG activity discriminated between different levels of cognitive demand during walking. Both behavioral parameters (e.g., blink duration or head motion) and blink-related EEG activity varied with walking conditions. Larger occipital N1 was observed during walking, relative to standing and traversing an obstacle course, which reflects differences in bottom-up visual perception. In contrast, the amplitudes of top-down components (N2, P3) significantly decreased with increasing walking demands, which reflected narrowing attention. This is consistent with blink-related EEG, specifically in Theta and Alpha power that, respectively, increased and decreased with increasing demands of the walking task. This work presents a novel and robust analytical approach to evaluate the cognitive demands experienced in natural work settings, which precludes the use of artificial task manipulations for data segmentation.

1. Introduction

In applied EEG research, there is a tradition of measuring the cognitive load that participants experience when they perform a real-world task and to understand how this load impacts information processing [1,2,3,4]. By observing changes in implicit EEG measures, inferences can be made about the corresponding changes in an individual’s mental state (e.g., alertness) [5], without interfering with natural work processes. Doing so provides an estimation of the cognitive demands of a particular task [6], which delivers insights into the task’s mental load structure. Nevertheless, such experiments are typically conducted in laboratory environments under highly controlled experimental conditions. While generalizable aspects of natural working environments can be emulated, they cannot be exactly replicated [7,8]. Recent advances in mobile EEG equipment now allow EEG to be robustly recorded in real world situations [9,10,11,12,13].

Walking is a common task in the real world. Contrary to popular belief, it is neither an involuntary nor an automatic process. Visual processing actively contributes to gait control [14,15], whereby the placement of each step must be identified and analyzed prior to the leg’s ballistic movement [15,16,17]. Specifically, a stable gait depends on ‘knowing’ the future location of the step and the upcoming surface foothold, during the second half of the preceding step [15,16]. In addition, walking speed [18,19] and posture [20,21] are strongly controlled by visual input. Thus, visual processing can be expected to adapt to the perceptual and cognitive demands of walking. For instance, increased attention is allocated to the lower visual field for gait control [22] and visual sensitivity increases in the periphery [23] to subserve the control of locomotion velocity. Finally, walking on complex ground is often accompanied by decreased head movements [24], presumably to stabilize gaze behavior. In spite of this, many studies on information processing during walking are restricted to walking on a flat indoor surface or a treadmill with constant velocity [23,25,26,27].

Some real-world EEG studies have indirectly verified the involvement of cognitive processes during walking [12,13,26,28,29] or while riding a bicycle [30,31]. Typically, these experiments are designed to require mobile and unrestricted participants to perform a cognitive task that is unrelated to their locomotion. Presumably, conflicts between the demands of the cognitive task and locomotion would result in an attenuation of event-related potentials (ERPs), or event-related spectral perturbations (ERSPs) evoked by the stimuli of the cognitive task. For instance, when participants were required to respond to an infrequent auditory tone, selective components of the ensuing ERPs/ERSPs were attenuated by locomotor activity [12,13,29,31]. To summarize, real-world locomotor activity places cognitive demands that indirectly diminish the amplitudes of the ERPs/ERSPs of unrelated auditory stimuli.

The above studies were executed using auditory tasks that were unrelated to gait control. However, Wickens’ limited capacity model [32] asserts that high interference between tasks should only be expected if their domains of resource demands overlap. This raises the question: Can the visual demands of walking be investigated using neurocognitive methods? To do so, it is necessary to refer to a discrete event that approximates the onset of relevant information, which is relied upon by the cognitive mechanisms (e.g., gait control) that underlie walking. Presenting such stimuli via additional devices (e.g., head-mounted displays), as is usual for auditory tasks, would introduce a tertiary task that is not directly related to the visual control of walking and, additionally, increase the risk of falling. It would also change the walking task.

To overcome this problem, the current work proposes naturally occurring blinks as such a discrete event that segments visual input. It is established that eye blinks do not occur randomly. Instead, they are functionally timed to segment periods of meaningful visual information. For instance, blinks tend to occur at the end of a sentence when reading [33], at the end of a scene when watching movies [34], or after a decision that has been made [35]. Interestingly, this behavior has also been observed during auditory information processing, suggesting that it might be a general trans-modal reflex [36]. Relying on these findings, we can assume that eye blinks mark a moment when the processing of a chunk of information is finalized. The re-opening of the eye, therefore, marks the moment when new input enters the eye, comparable to the presentation of an external stimulus. Thus, event-related EEG that is time-locked to the blink could be treated similarly to event-related activity evoked by an external stimulus. Indeed, some studies have demonstrated that blink-evoked potentials are modulated by the cognitive demands of everyday tasks, either in a work-place simulation [4] or while a pedestrian navigates in a city [11].

To summarize, eye blinks can serve as discrete events that could allow us to directly estimate the cognitive load of a dual-task walking situation from both ERPs and ERSPs. Blinks mark the onset of incoming visual information by which processing should be affected depending on the level of gait complexity. To verify this, we re-analyzed two existing data sets. In both experiments, participants had to either stand, walk in a meadow, or walk across an obstacle course while performing auditory tasks [12,13]. Blinking behavior and blink-evoked EEG-activity were analyzed for both experiments.

Blinks were first identified as independent components in the EEG data, decomposed using an independent component analysis (ICA). Using Gaussian-fit-based template matching (see Section 2), individual eye blinks were identified from the relevant independent component and their peak amplitudes served as epoch markers for ERPs and ERSPs. The following predictions were evaluated in the epoch-averaged data. First, walking on more demanding ground should demand enhanced attentional allocation that would be reflected in larger amplitudes in the posterior N1 and parietal P2 components. The same can be expected for fronto-central N2, a known correlate of executive control that is sensitive to task demands [37]. Additionally, the parietal P3, which is similarly connected to cognitive resource allocation, typically decreases with increasing task demands [38]. Highly complex visual input during walking is predicted to generate similar effects. Some studies suggest that high scene complexity could decrease the amplitudes of N1, N2, and P3 amplitude [39], indicating attentional narrowing towards relevant information and the suppression of irrelevant information. The former predictions could reflect the role of elementary processes during walking, while the latter could highlight the systemic aspects of information processing. Both sets of predictions are evaluated in our analyses of eye-blink-evoked ERPs.

The same set of predictions may hold true for ERSPs, in particular, Theta and Alpha activity. A purely function-based view would predict that more demanding tasks reduce levels of Alpha power, since more task engagement is involved [40,41], However, Alpha activity could also increase with more complex scenes [42]. Additionally, this effect can be interpreted in terms of increased inhibitory processing for irrelevant information in favor of relevant signals.

2. Methods

2.1. Participants

The datasets used to calculate blink-evoked potentials were taken from two studies that were performed between June and November 2018. From these, 36 participants were included in the calculations presented here (18 female, 18 male). Four participants were excluded, as the blink-detection algorithm was unable to accurately quantify their blinking behavior. The participants’ ages ranged from 19 to 30 years (M = 23.44; SEM = 0.51). For active participation in the study, participants received a monetary compensation of EUR 10 per hour or course credit. Both studies were approved by the local ethics committee of the Leibniz Research Centre for Working Environment and Human Factors and were conducted in accordance with the Declaration of Helsinki. Prior written consent to participation was obtained from all participants.

2.2. Tasks

2.2.1. Auditory Tasks

Both studies analyzed used well-known auditory paradigms. All stimuli were presented using passive noise-canceling in-ear headphones (Bose QuietComfort 20, Framingham, MA, USA). In the first study [12], participants were presented with an auditory oddball task in which they had to react to a target sound while ignoring the more frequent standard sounds. Per condition block, they were presented with a continuous stream of 450 low-pitched (600 Hz “standard”) or high-pitched (900 Hz “deviant”) sinusoid tones presented in randomized order. Participants had to press a response button when presented with deviant tones and had to withhold their response for standard tones.

The second study [13] used a cued auditory task-switching paradigm. Here, participants were presented with an auditory cue—either a low 600 Hz or a high 900 Hz sinusoid tone—before hearing a spoken German number between one and four or six and nine. Depending on the cue-tone, participants had to indicate by a left- or right-hand button press whether (a) the presented number was lower or higher than five or (b) the presented number was odd or even using manual response buttons held in the left and right hand. There were two possible conditions: a 5 min repeat task (96 trials, either with a lower/higher or odd/even decision), and a 10 min switch task exercise (208 trials) while executing one of the three motor complexity conditions (see next paragraph).

2.2.2. Walking Conditions

In both studies analyzed, participants had to perform three different motor complexity conditions on the outside premises of the institute. Here, they had to (a) stand still in a specific place at the corner of the obstacle course that was built for the two studies. In the other conditions, they had to (b) walk around the outside of the obstacle course at a comfortable speed, as many laps as they managed in the time, or to (c) walk and traverse the obstacle elements. This obstacle course consisted of two 4-step staircases, two balancing beams, and two balancing boards with hole cut-outs (see Figure 1). The duration of all three conditions was equivalent to the presentation time of the respective auditory tasks (study 1: 15 min; study 2: 5 min repeat, 10 min switch). The locomotor conditions (walking, and walking with obstacle elements) were executed in a self-paced manner.

In both studies, all participants had a total experimentation time of 90 min, with 30 min spent on each movement complexity condition (15 min of standing, walking, and obstacle course walking, both in clockwise and anti-clockwise direction). The order of cognitive walking conditions was quasi-randomized using a Latin-square design. Thus, each walking condition was performed in the first and in the second half of the experiment.

2.3. Recording and EEG Data Processing

2.3.1. Data Recording and Preprocessing

The EEG was recorded using an EEG cap with 30 active electrodes in an adjusted 10-10 system montage (Fp1, Fp2, F3, F4, F7, F8, Fz, FC1, FC2, FC5, FC6, C3, C4, Cz, T7, T8, CP1, CP2, CP5, CP6, P3, P4, P7, P8, Pz, PO9, PO10, O1, O2, Oz). FCz was used as the online reference and AFz served as the ground electrode. All electrodes were prepared using SuperVisc electrolyte gel (Easycap GmbH, Herrsching, GER) to reach an impedance of 10 kΩ or less. Data were recorded using a 32-channel LiveAmp (Brain Products GmbH, Gilching, GER) with a sampling rate of 500 Hz and a bandwidth from DC to 131 Hz (3rd order sinc filter, −3 dB). Data were stored on the microSD card inserted into the device.

The amplifier provided additional data from three accelerometers (x, y, and z-axis, 12-bit, range +/− 2 g). For each accelerometer data point, the center-of-gravity vector was calculated as cgv = sqrt(x² + y² + z²). Thereafter, data were divided into 1s segments and the averages of head acceleration and of the change in head acceleration (head dynamics) were calculated for each segment.

After removal of the accelerometer channels, continuous EEG data were band-pass filtered (0.1–40 Hz) and checked for gross artifacts using the EEGLAB function pop_rejcont, with a frequency range for thresholding between 20 and 40 Hz. The frequency upper threshold was set to 10 dB with an epoch length of 0.5 s. The number of contiguous epochs necessary to label a region as artifactual was 4. Once a region of contiguous epochs was labeled as artifact, 0.25 s additional trailing neighboring regions on each side were added. Remaining data were submitted to the PrepPipeline [43] including the rejection of corrupt EEG channels based on a robust average reference.

This cleaned data set was high-pass filtered at 1 Hz and down-sampled to 250 Hz for ICA decomposition. Epochs of 1 s (based on the segments defined for head motion parameters) were extracted, checked for artifacts, and entered into an ICA. The ICA solution obtained from this data was applied to the original cleaned continuous 500 Hz data set with the initial band pass filtering (0.1–40 Hz).

Blinks were determined based on the ICA solution. The blink IC was selected automatically by correlating the time course of the average of anterior EEG activity with each IC in the activation matrix. The IC with the highest correlation was used for further analyses. Throughout the time course of the blink, related IC local maxima were searched. For each of these local maxima, a Gaussian curve was fitted in a time window +/− 120 ms around the peak. Only those blink event candidates with a sufficient Gaussian wave form (goodness of fit (r-squared) > 0.9) and an amplitude that exceeded 75% of the median amplitude of all peaks were included in the further analyses as valid blinks. The amplitude criterion was selected to exclude rapid vertical eye movements and incomplete blinks. Each blink was additionally labeled with the amount of head acceleration and head dynamics at the moment of its occurrence.

After this procedure, ICs with less than 30% brain activity, as classified by ICLabel, [44] and more than 30% eye activity were removed from the data. Cleaned continuous data were segmented from −600 to 1800 ms around the blink maximum with a baseline from −400 to −200 ms relative to the blink maximum. The obtained segments were again entered into an automatic segment rejection procedure using EEGLAB function pop_autorej, with a voltage threshold of 500 µV. The probability threshold for detection of improbable data was set to 5 standard deviations. The maximal 10% of total trials were rejected per iteration. The remaining segments were considered as valid trials, which were then entered into further analyses. All analyses, as well as behavioral and mental state analyses, were based on the same artifact-free blink trials.

2.3.2. Data Analyses

Before entering the description of the analyses in detail, two principal decisions regarding the design must be mentioned. First of all, the three walking conditions were not treated as a continuum. If it can be assumed that walking on a meadow reflects a largely automated behavior, then the comparison between standing and walking on a meadow mostly addresses the factor of locomotion. When comparing walking on a meadow and walking across the obstacle course, it can be assumed that these two conditions vary mostly with respect to the level of cognitive control of locomotion. Thus, these two comparisons were statistically tested separately for all statistical analyses in order to not intermix these two different constructs. Secondly, the differentiation between the two auditory stimulation tasks (experiment 1: oddball; experiment 2: task switch) as a between-participant factor was abandoned after pre-analysis of the data, since hardly any impact of this factor was observed on behavioral and EEG data.

For each eye blink, identified as described above, its duration and its temporal distance from the preceding blink were identified. Based on the distribution of data, blink-intervals between 400 and 5000 ms were rated as valid to ensure proper separation of blinks, as well as the inclusion of spontaneously occurring blinks only. Data obtained for both measures were entered into two repeated measure ANOVAs each, comprising the within-participant factors walking condition (Analysis 1: standing vs. walking on meadow; Analysis 2: walking on meadow vs. walking across the obstacle course), and block (1st vs. 2nd run).

The parameters of head motion—namely, head acceleration as the calculated average of the center of gravity of head acceleration, as provided by the gyro-sensors of the EEG amplifiers in the 1s segment in which the blink occurred, and the change in this parameter in time (head dynamics; first derivation of the former)—were entered into the same analyses.

A common sequence of visual ERP components could be observed in the grand averages across all participants (see Figure 2).

The amplitudes of all ERP components analyzed were determined as the mean amplitude +/− 20 ms around the ERP component maximum in the grand average across all conditions. Following an occipital (O1, Oz, O2) N1 at 120 ms, a parietal P2 at Pz was observed, which peaked at 150 ms. At anterior sites (Fz, Cz), an N2 (300 ms) was observed, and at the same time (274 ms), with similar dissociation between conditions, an occipital P2 was also observed. A P3 component was found at Pz and peaked at 254 ms. Mean amplitudes for all these components were entered into repeated measure ANOVAs with the same factors as for the behavioral parameters.

ERSPs were obtained by convoluting the data with complex Morlet wavelets. A set of 30 wavelets was used with linearly spaced frequencies ranging from 2 to 30 Hz. The wavelets were constructed in a way in which the full length at half maximum ranged from 500 ms for the lowest frequency to 100 ms for the highest frequency. The resulting ERSPs were decibel-normalized using a time window from −500 ms to −200 ms, relative to the blink maximum, as baseline. Frequency-band-specific activity was calculated by averaging the ERSPs in the Theta (3.5–6.5 Hz) and Alpha (8–13 Hz) ranges for 4 electrode pools over the frontal (Fz, FC1, FC2, Cz), temporal (C3, C4, FC5, FC6), parietal (CP1, Pz, CP2) and occipital (O1, Oz, O2) sites. Base activity was determined as the mean power between 600 and 1000 ms after the blink maximum, since the time preceding blink execution might be corrupted by preparatory processes. Spectral perturbations were defined as the power change from base activity to the phasic maximum following the blink (Theta: 80–120 ms; Alpha: 120–160 ms). These values were also entered into separate repeated-measures ANOVAs for the comparison of standing vs. walking on meadow and walking on meadow vs. walking across the obstacle course, with experimental block and electrode pools as additional factors.

Post hoc tests were performed when indicated and permissible. If required, p-values were Greenhouse–Geisser corrected. Effect sizes are given as adjusted partial eta squared [45].

3. Results

3.1. Behavioral Data

For the sake of brevity, only statistically significant results and the comparison between the analysis conditions will be given below. For all the listings of all the statistical analyses performed, please refer to Appendix A.

Head acceleration (see Figure 3) was greater when participants walked on the meadow compared to standing (F(1,35) = 144.81, p < 0.001, _adjη_p² = 0.80) and further increased when the obstacle course was used (F(1,35) = 195.68, p < 0.001, _adjη_p²= 0.84). The interaction of walking condition and block when the obstacle course and meadow are compared (F(1,35) = 8.01, p = 0.015, _adjη_p² = 0.16) indicates that head acceleration was highest when participants passed across the obstacle course the first time. Additionally, head dynamics naturally increased when walking in the meadow compared to standing (F(1,35) = 330.06, p < 0.001, _adjη_p² = 0.90); however, when comparing the two walking conditions, head dynamics were reduced when walking across the obstacle course (F(1,35) = 9.94, p = 0.006, _adjη_p² = 0.20).

Blink interval did not differ between walking and standing (F(1,35) = 0.12, p > 0.1) but decreased significantly when the obstacle course was mastered compared to walking on the meadow (F(1,35) = 46.06, p < 0.001, _adjη_p² = 0.56). Blink duration was longer when standing compared to walking on the meadow (F(1,35) = 9.23, p = 0.009, _adjη_p² = 0.19). In this analysis, an effect of experimental block was also visible (F(1,35) = 6.79, p = 0.027, _adjη_p² = 0.14). In the comparison of the two walking conditions, meadow vs. obstacle course, no significant effect was found (F(1,35) = 3.87, p > 0.1). However, again, an effect of block was found (F(1,35) = 9.47, p = 0.008, _adjη_p² = 0.19) with longer blinks in the second block.

3.2. EEG Parameters

3.2.1. ERPs

ERP waveshapes were substantially different between walking conditions but fairly similar across blocks (see Figure 2). Most of the parameters neither revealed a main effect of experimental block nor an interaction between walking conditions and experimental block, with all F-values being around 1 and below. Thus, they will not be mentioned in detail here, again, for the sake of brevity and with all results being given in Appendix A.

The amplitude of the N1 was increased when walking compared to standing (F(1,35) = 14.72, p = 0.001, _adjη_p² = 0.28), but decreased when traversing the obstacle course was compared to walking on the meadow (F(1,35) = 47.57, p < 0.001, _adjη_p² = 0.56) (see Figure 4 for the topography of this effect, which is clearly focused in visual areas). For the comparison of meadow vs. obstacle course, an interaction of block by walking condition was found (F(1,35) = 7.24, p = 0.021, _adjη_p² = 0.15), indicating that the occipital N1 amplitude was highest when participants walked on the meadow for the second time.

In contrast to this finding, the fronto-central N2 amplitude decreased from standing to walking (F(1,35) = 14.53, p = 0.001, _adjη_p² = 0.27), and further from walking the meadow to walking the obstacle course (F(1,35) = 16.77, p < 0.001, _adjη_p² = 0.30).

The parietal P2 slightly increased in amplitude from standing to walking the meadow (F(1,35) = 8.44, p = 0.012, _adjη_p² = 0.17). The continuation of this effect, visible in Figure 4, from walking the meadow to walking the obstacle course, did not reach significance (F(1,35) = 3.17, p > 0.1).

The occipital P2, however, showed a highly significant decrease in amplitude with increasing complexity of the walking conditions: standing vs. walking the meadow (F(1,35) = 15.28, p < 0.001, _adjη_p² = 0.28), and walking on meadow vs. walking the obstacle course (F(1,35) = 25.57, p < 0.001, _adjη_p² = 0.41).

Finally, the parietal P3 amplitude showed a linearly decreasing effect of increasing walking complexity. The P3 amplitude was slightly larger for standing compared to walking (F(1,35) = 5.24, p = 0.056, _adjη_p² = 0.11) and decreased further when participants walked the obstacle course ( F(1,35) = 16.72, p < 0.001, _adjη_p² = 0.30). For the P3 component, however, significant effects of block were found in both analyses (F(1,35) = 10.28, p = 0.006, _adjη_p² = 0.20; and F(1,35) = 6.15, p = 0.036, _adjη_p² = 0.13), with larger amplitudes in the second blocks. In the latter analysis, comparing walking on meadow vs. walking the obstacle course, an additional interaction of walking condition by block was obtained ( F(1,35) = 8.87, p = 0.011, _adjη_p² = 0.18), likely due to the reduction of the block effect on P3 amplitude in the obstacle course condition.

3.2.2. ERSPs

Theta: The Theta base level (see Figure 5) decreased for standing compared to walking on the meadow (F(1,35) = 68.13, p < 0.001, _adjη_p²=.65). Furthermore, an interaction of walking condition by channel pool (topography) was observed (F(3,105) = 16.21, p < 0.001, _adjη_p² = 0.29). This was due to the effect of walking being most pronounced at occipital sites, both numerically and with respect to statistical power (frontal: F(1,35)=43.58, p < 0.001, _adjη_p² = 0.54; temporal: F(1,35) = 32.63, p < 0.001, _adjη_p² = 0.47, parietal: F(1,35)=27.51, p < 0.001, _adjη_p² = 0.42; occipital: F(1,35) =132.97, p < 0.001, _adjη_p² = 0.79). The Theta base level additionally increased with block (F(1,35) = 10.20, p = 0.006, _adjη_p² = 0.20). This effect was strongest at occipital sites as well (frontal: F(1,35) = 2.18, p>.1; temporal: F(1,35) = 4.96, p = 0.064, _adjη_p² = 0.10; parietal: F(1,35)=4.39, p = 0.086, _adjη_p² = 0.09; occipital: F(1,35) =12.19, p = 0.002, _adjη_p² = 0.24).

When the two walking conditions are compared, neither walking condition nor block modulated Theta base power.

Theta perturbations were lower in power for walking on the meadow compared to standing (F(1,35) = 16.22, p < 0.001, _adjη_p² = 0.30), an effect which was modulated by topography (F(1,35) = 25.60, p < 0.001, _adjη_p² = 0.41), with the effect of walking being clearly visible only at temporal and occipital sites (temporal: F(1,35) = 13.41, p < 0.001, _adjη_p² = 0.26; occipital: F(1,35) = 36.54, p < 0.001, _adjη_p² = 0.50), but not at frontal or parietal sites (frontal: F(1,35) = 2.98, p > 0.1; parietal: F(1,35) = 0.50, p > 0.1). Though the main effect of factor experimental block did not reach significance (F(1,35) = 1.87, p > 0.1), a slight decrease in evoked Theta power was observed in the second block while standing (interaction of walking condition x block: F(1,35) = 7.52, p = 0.019, _adjη_p² = 0.15).

Additionally, when the two walking conditions are compared, Theta perturbations decreased with walking complexity (F(1,35) = 31.28, p < 0.001, _adjη_p² = 0.46). This effect was most pronounced at the frontal and occipital leads (interaction of topography x walking condition: F(3,105) = 10.55, p < 0.001, _adjη_p² = 0.21; frontal: F(1,35) = 10.41, p = 0.005, _adjη_p² = 0.21; temporal: F(1,35) = 10.77, p = 0.021, _adjη_p² = 0.21; parietal: F(1,35) = 4.79, p = 0.071; occipital: F(1,35) = 33.37, p < 0.001, _adjη_p² = 0.47).

Alpha: The Alpha base level (see Figure 6) did not differ for the comparison of standing vs. walking on the meadow (F(1,35) = 1.74, p > 0.1), but an interaction of walking by topography was found (F(3,105) = 38.43, p < 0.001, _adjη_p² = 0.51), due to the fact that a difference between standing and walking in Alpha power level was observed at occipital sites only (frontal: F(1,35) = 0, p > 0.1; temporal: F(1,35) = 0.54, p > 0.1; parietal: F(1,35) = 3.59, p > 0.1; occipital: F(1,35) = 32.40, p < 0.001, _adjη_p² = 0.47). The Alpha level also increased with block (F(1,35) = 12.67, p = 0.002, _adjη_p² = 0.24), and this effect differed across electrode sites as well (interaction of block x topography: F(3,105) = 5.94, p = 0.015, _adjη_p² = 0.12; frontal: F(1,35) = 27.05, p < 0.001, _adjη_p² = 0.42; temporal: F(1,35) = 9.38, p = 0.008, _adjη_p² = 0.19; parietal: F(1,35) = 13.94, p = 0.001, _adjη_p² = 0.26; occipital: F(1,35) = 0.44, p > 0.1). Finally, a triple interaction of walking condition by block by topography was observed (F(3,105) = 3.87, p = 0.032, _adjη_p² = 0.07). This modulation of the block effect across channel topography was observed only for walking on the meadow (F(3,105) = 9.83, p < 0.001, _adjη_p² = 0.18), but not for standing (F(1,35) = 1.62, p > 0.1).

Additionally, in the analysis across the two walking conditions, the Alpha base level further increased when traversing the obstacle course (F(1,35) = 14.12, p = 0.001, _adjη_p² = 0.27). This effect was, again, most pronounced at occipital sites (interaction of walking condition x topography: F(3,105) = 21.79, p < 0.001, _adjη_p² = 0.37; frontal: F(1,35) = 3.63, p > 0.1; temporal: F(1,35) = 3.75, p > 0.1; parietal: F(1,35) = 0.93, p > 0.1; occipital: F(1,35) = 33.91, p < 0.001, _adjη_p² = 0.48). As in the comparison of walking on the meadow vs. standing, the Alpha level also increased with block (F(1,35) = 6.30, p = 0.033, _adjη_p² = 0.130, which was least pronounced occipitally (interaction of experimental block by topography: F(3,105) = 12.76, p < 0.001, _adjη_p² = 0.25; frontal: F(1,35) = 8.79, p = 0.011, _adjη_p² = 0.18; temporal: F(1,35) = 4.58, p = 0.008, _adjη_p² = 0.09; parietal: F(1,35) = 23.05, p < 0.001, _adjη_p² = 0.38; occipital: F(1,35) = 2.08, p > 0.1).

No main effect of walking was observed for Alpha perturbations when standing and walking on the meadow were compared (F(1,35) = 0.39, p > 0.1). However, we found a significant interaction of walking conditions and topography (F(3,105) = 20.70, p < 0.001, _adjη_p² = 0.35). An effect of walking condition was only observed at the parietal electrode cluster (frontal: F(1,35) = 0.11, p > 0.1; temporal: F(1,35) = 0.03, p > 0.1; parietal: F(1,35) = 23.93, p < 0.001, _adjη_p² = 0.39; occipital: F(1,35) = 5.43, p = 0.051, _adjη_p² = 0.11). Alpha perturbations marginally decreased with block (F(1,35) = 4.78, p = 0.071, _adjη_p² = 0.10). This effect differed between walking conditions (F(1,35) = 5.23, p = 0.056, _adjη_p² = 0.11). Only for standing (F(1,35) = 6.93, p = 0.025, _adjη_p² = 0.15), but not for walking (F(1,35) = 0.01, p > 0.1), was a block effect observed.

In the comparison of walking on the meadow versus walking over the obstacle course elements, none of the experimental factors modulated Alpha perturbations (all p-values > 0.1).

4. Discussion

Locomotion interferes with cognitive processing, even when walking on flat surfaces that elicit highly automated gait patterns [29,46]. Walking on uneven surfaces requires visual control of stepping, which further increases interference. Visual analysis of the ground surface and step-planning predetermine the location and footholds, even before the ballistic motion of the leg is initiated. Thus, the efficient interaction of vision and motor planning is a central factor for safe locomotion [15,17,22].

It is not an easy endeavor to investigate the interaction of vision and walking on natural ground. This is especially true when neurocognitive methods are applied to obtain an in-depth insight into mental states and processing. Using additional visual stimuli (e.g., presented via data glasses) while walking could add another task that may interfere with visual processes. Such a task—when referring to the multiple resource theory by Wickens [32]—might impair safe walking by competing for resources that are necessary to detect safe walking paths. We believe, like others, that it is preferable to avoid walking-unrelated stimulation, if the goal is to investigate the load of walking on the visual system [4,47]. Nevertheless, repetitive, discrete stimulation are required to time-lock and segment EEG activity for event-related analyses [48]. Here, we propose that eye-blink activity can serve as time-discrete event markers instead.

Besides moisturizing the eye, naturally occurring blink-events have been shown to segment the flow of continuous visual information [33,34,35]. Thus, EEG data that are blink-segmented are likely to contain event-related cognitive activity related to visual information segmentation. This removes the need to introduce additional tasks typically used by workload measurement studies.

We applied this method of blink-related EEG activity to data from two recently published studies [12,13]. These studies were originally performed to investigate how tasks with increasing motor complexity (i.e., standing, walking on meadow, and walking while traversing an obstacle course in a natural, outside environment) interfered with auditory information processing. In the studies that the analyzed data were taken from [12,13], clear behavioral effects were shown in auditory tasks, but not the expected differentiation of ERP effects between walking conditions. There are at least two reasons for this outcome. First, walking on a meadow is not a highly automated task, since the meadow’s surface is uneven and, therefore, requires high levels of cognitive control for safe locomotion. Thus, it is more comparable to traversing an obstacle course than walking on a completely flat floor (e.g., in a hallway). Alternatively, the interference between auditory information processing and walking demands might be negligible, since the functional overlap of visual movement planning and auditory task execution is not high. Thus, only the act of locomotion might affect auditory processing, but not visual control.

Consistent with the assumption of the specific interaction between visual processing and walking, the effects found in the analyses presented herein can be split into three main effect categories: the effects of locomotion (effects that only differ between standing and general locomotion), the effects of walk load (steadily inclining effects from standing to walking on meadow to traversing the obstacle course) and, finally, the effects of experimental block (indicators for reliability and for time-on-task effects). Additionally, a few parameters showed specific adaptation to the cognitive processes involved that did not follow the systematic categorization of effects as outlined above.

Changes in the parameters over time are indicators of data quality on the one hand, but are also possible indicators for time-on-task effects. For both the ERP and the time frequency data, high similarity of the blink-evoked responses was obtained across blocks. This was found for both the waveshape morphology and the sensitivity of the different measures to the experimental conditions. Strong time-on-task effects were only observed for the duration of blinks that increased in later parts of the experiment. Block modulation of the Alpha base level was restricted to walking on the meadow. At this point, it must be noted that, in contrast to the rather monotonous experiments in which mental fatigue has been investigated thus far, complex behavior in a natural surrounding does not lead to decreased engagement and, consequently, decreased motivation. Thus, no signs of mental fatigue were found, despite the extended duration of the experiment. The lack of a modulation of parameters across blocks before most indicate the high reliability of the measures extracted.

Blink duration was reduced in locomotive states, relative to standing still. This suggests increased visual information processing while walking [49,50]. Additionally, the parietal P2 amplitude and Theta base level were increased, which would be expected for stronger attentional allocation [51,52]. This indicates that more effort had to be invested to filter out relevant information during locomotion [30,53].

Looking into effects of walk load, the decrease in amplitude of the fronto-central N2, the occipital P2, and the P3 may indicate the reduced availability of mental resources when walking conditions become more demanding [13]. However, the effects in the N2 and the P3 may also be interpreted in terms of attentional narrowing when processing complex input [42]. The same interpretations can be applied to the decrease in Theta perturbations. There are also results that do not follow the hypothesized directions of effects. Both head dynamics and the occipital N1 were maximal when walking on the meadow. The increase in these parameters when compared to the standing condition can be explained by locomotion. Increases in head motion and visual attention must be applied to succeed in the task. The inversion of the effects when traversing the obstacle course can be explained as outlined in the introduction. Aspects of head motion tend towards stabilization when walking conditions become more complex [24]. Such a behavior might help to increase eye fixation stability for the elevated step planning demands in obstacle course conditions [15,16]. In this situation, attention is both focused and narrowed to the information of interest. Considering the modulation of the N1 topography, we see a central contribution of the visual system. In the beginning, the narrow spot indicates a contribution of early visual areas, whereas the spread after 120 ms indicates the propagation towards higher-order areas [54,55].

Such attentional narrowing is consistent with the observations of differences in the fronto-central N2 and other components, and is strongly supported by the modulation of Alpha activity. From laboratory data, it is well-known that Alpha power decreases with task complexity [56,57]. This effect has been attributed to attentional withdrawal or the internalization of attention for easy tasks [58]. The inversion of this effect indicates that parts of the attentional system have been idled to receive only relevant information.

In summary, blink-related EEG activity provides meaningful information about mental load and visual processing during walking. Although we found characteristics of the blinks themselves (e.g., duration) and behavioral measures—such as head motion—to indicate specific cognitive states, electrophysiological-event-related components were found to be even more insightful. While the occipital N1 was sensitive to the selective adjustment of visual demands, changes in later and more cognitively driven components (N2, P3) indicated an overall narrowing of attention. This was also evident in Alpha and Theta power, with functionally plausible patterns of increases and decreases in line with the respective walk load. While the results of the stimulus-locked EEG analyses in the original studies [12,13] did not unveil differences in the cognitive processing of auditory stimuli with increasing walk load, blink-related analyses allowed us to specifically look into visual perceptive and cognitive resource-management processes not bound to the auditory task of the original paradigms, without imposing a manipulated event-structure on the participants.