Walkers use step synchronization as a form of non-verbal social communication

0
1005

Walking is one of our most natural, daily actions.

Now, a new study led by a Tohoku University researcher suggests that walkers use step synchronization as a form of non-verbal social communication.

The results lend credence to the effects of psychological traits on movement interaction between humans.

Published in PLOS ONE, the study results demonstrate how people’s traits and first impression affect their nonverbal communication, i.e. synchronous walking.

In order to conduct the study, researchers divided participants into groups. In total, ten single-gender groups were formed – five female and five male.

Group members took turns being paired up with other members, and they walked together along a quiet, barrier-free path.

They wore voice recorders, and motion sensors disguised as GPS devices recorded their walking movements.

Researchers carried out the experiment under three conditions: a half silent walk half conversation condition where participants did not speak for half of the journey, yet conversed on the way back; a silent walk condition where participants did not converse for the entirety of the journey; and lastly, a non-walking condition where participants did not walk and sat quietly filling in a questionnaire in a classroom.

Participants had no prior knowledge of each other, and were asked to rate their impression of their partners before and after each walk using the interpersonal judgment scale (IJS). Furthermore, researchers misled participants about the true nature of the study to prevent them from intentionally synchronizing their steps.

The results revealed an increase in the impression ratings for the two groups of participants who walked together, but not for the group of participants who simply spent time together.

This suggests that walking side-by-side, even without verbal communication, is sufficient to alter the social relation between two strangers.

Results also showed that conversations further enhanced impressions for participants who were allowed to talk.

Ultimately, the experiment successfully dissociated the contribution of verbal communication from walking step synchronization, which was inseparable in previous studies.

Female pairs, compared to male pairs, exhibited higher walking synchrony in this experiment.

The researchers also found that pairs with a better first impression had greater synchronization in their steps – particularly for female participants.

In addition to social relation, personal traits are also important. Female pairs, compared to male pairs, exhibited higher walking synchrony in this experiment.

There is also an age effect – older participants tend to synchronize with their partners more in walking. Participants with lower autistic tendencies synchronize better than pairs of higher autistic tendencies.

“It is very surprising for us to discover that a person’s traits and our first impressions are reflected in the subtle action of walking. I think most people are not even aware that their steps are synchronized with other people as they walk,” said Dr. Chia-huei Tseng, an associate professor at the Research Institute of Electrical Communication (RIEC) at Tohoku University.

“It was previously known that a person’s physical parameters such as height and weight affect how their movements interact with others. Now we know psychological traits also have an effect.”

“There is a growing awareness of the validity of interpersonal interaction research in real-world scenarios, but daily natural environments are rich in their contextual information, making experiment control a challenge,” said Dr. Miao Cheng, a post-doc researcher at NTT Communication Science Laboratories.

“Our study is important because it is an approach to make the most use of an ecological paradigm while using scientific experimental method to control possible confounding variables to examine the function of implicit body synchrony.”


To successfully navigate any environment, a walker must adapt to the surface they walk on, avoid obstacles, change speed, and plan movements according to their goals [1,2]. Processes that impact walking behavior range from peripheral processes to high-level decision-making processes [3,4].

Walking, like all movement, often takes place in the context of other people. Humans, when walking, must often generate their actions according to the movements of people around them.

Coordination of movements with others during everyday activities helps to achieve shared goals efficiently and fluently, for example when playing sports, moving an object together, or walking side-by-side. In many cases, synchronization is needed to successfully carry out the task at hand.

However, evidence from laboratory experiments suggests that synchronization between people does not only take place when it is required by the task. There is evidence for uninstructed movement coordination from laboratory experiments where subjects are asked to perform various artificial tasks, for example, sway pendulums, sit in rocking chairs, or spontaneously move their arms [511].

Coordination across individuals thus appears to be fairly ubiquitous, occurring even when it is not required by the task and there is no instruction to do so.

Previous laboratory-based studies have investigated if people walking side-by-side synchronize their movements. Walker synchronization is common when people walk side-by-side, and has been found in over-ground walking studies [4,1214] and in treadmill walking studies [1520].

For example, Zivotofsky and Hausdorff (2007) found that when people were asked to walk side-by-side, they walked in anti-phase (left-right, right-left), with the left leg of one walker aligned with the right leg of the other walker, without instruction. van Ulzen et al. (2008) tracked subjects during side-by-side treadmill walking and found both in-phase (left-left, right-right) and anti-phase (left-right, right-left) walking with preferred synchronization modes. In sum, previous results provide evidence that pairs tend to walk in phase or in anti-phase.

However, as with many movement science paradigms, it is not known if walker synchronization occurs in real world settings, outside of the laboratory.

This is especially relevant for the walking studies in the laboratory: researchers have found a substantial difference in biomechanics between walking on the treadmill and walking on level ground [21,22]. The synchronization of pairs of walkers observed on treadmill might not generalize to real life settings.

There have been previous examinations of walker synchronization during over ground walking in laboratory settings. Most laboratory studies used motion capture systems applied to the limbs to quantify walking parameters. Research relied on reviewer ratings to address walking over ground in one study [14].

In a second lab-based study of over ground walking, accelerometers were attached to the trunk, and although their findings suggest synchronization between walkers, it is not possible to conclude whether walking is in phase or in anti-phase [23]. The general lack of work using motion capture outside of the lab is understandable since motion capture technologies were either low in precision, like smartphone sensors, or low in portability, with previous experiments requiring that subjects wear markers or recording equipment [24,25].

This made motion capture a challenge for studying walker synchronization in real life. Thanks to recent work on pose estimation from 2D video, it is now possible to perform marker-less pose estimation with only 2D video data as input [2628].

These algorithms take as input a 2D RGB image and output an estimate of human pose in image coordinates with reasonable accuracy. Marker-less pose estimation applied to videos provides the opportunity to ask questions about naturalistic behavior outside the laboratory.

Here we apply pose estimation to online videos to ask how people synchronize their movements when they walk side-by-side. We used a state-of-the-art pose-estimation algorithm, OpenPose, to extract human body pose from 2D videos [26].

OpenPose finds joint positions of people in videos and fits a 2D skeleton model to each person in the image. We built a simple search algorithm to find video segments containing pairs of people walking side-by-side in YouTube videos. From the resulting video segments, we extracted the pose of people in the video.

In order to track person identity through videos, we used our own simple implementation of a tracking algorithm that minimized the distance between pose estimates across frames. We then examined pose estimates to ask if people walking together synchronized their movements, using vertical ankle displacement to quantify walking angle and thus phase.

Results

We asked if people synchronize their walking when they walk side-by-side in naturalistic settings. To do so, we analyzed videos found on YouTube. Within YouTube videos, we searched for video segments with pairs of people walking. From video segments, we extracted the pose of each member of the walking pair.

In order to examine walker synchronization, we analyzed displacement between left and right ankles for each member of the pair. Based on the displacement signal, we extracted the walking frequency and mean relative phase for each pair to examine walker synchronization.

To find relevant videos, we searched for videos on youtube.com using the search term ’walking in’ followed by the names of major cities. Our search resulted in 363 videos. We excluded 113 videos because they did not include footage of pairs of people walking.

We excluded 48 videos because of large amounts of camera movement, cropped video frames, camera angles which prevented pose estimation, and poor visibility.

We excluded 3 videos because the format of the video was not suitable (a large amount of occluding text on the screen and panoramic videos). After screening, we included a total of 199 videos, which gave us a reasonably sized dataset for quantifying synchronization.

Within these videos, our algorithm identified 888 video segments with pairs of people walking lasting at least 2.5 seconds.

Some segments were unsuitable because

1.) the segment did not contain pairs of adults walking continuously towards or away from the camera in well-lit conditions (258 segments),

2.) the segment included large amounts of camera movement, occlusion, or cropping of body parts from the video which prevented successful pose-tracking (197 segments). In the end, we extracted pose from the 433 suitable video segments containing 441 pairs of walkers, among which pose estimation or tracking failed for 93 walking pairs.

Thus, our final sample for walking analyses include 348 video segments with a mean duration of 4.73 seconds (SD = 5.99, Fig 3). Walkers in the video had a mean age rating of 30.66 years (SD = 12.34). Although most walkers were in the young age range of 20~40 years (71%), we still had 29% of walkers in the range of 40 to 70 years. Thus, we have a reasonably representative dataset to investigate our questions.

We wanted to quantify the degree of synchronization in the walking phase of each pair. To do so, we examined the phase relationship between two oscillatory walking patterns, using the signed y-displacement between left and right ankles.

From the time series of left and right ankle positions in y-coordinates (Fig 4A and 4B), we extracted the cyclic motion of walking for each member of a walking pair. We first detrended the y-coordinates of the left and right ankles (Fig 4C nad 4D). Then we normalized and low pass filtered the detrended signal (Fig 4E and 4F). For each walker, we computed the displacement between the left and right ankles (Fig 4G), then computed the Hilbert phase of the displacement time series (Fig 4H).

Finally, we computed the relative phase between Walker 1 and Walker 2. We examined the distribution of the relative phase to quantify walker synchronization (Fig 4I). The representative video segment in the Fig 4 showed a predominant in-phase synchronization as the relative phase clustered around zero. The ability to estimate relative phase from videos enables us to estimate synchronization behavior in the population.

Fig 4

Fig 4Extracting relative phase from the image y-coordinates of the left and right ankles.

We wanted to validate the outputs from pose estimates. We thus compared the walking frequency and mean relative phase computed from pose estimates with those extracted from a ground truth signal. We found reasonable agreement between the ground-truth estimates and pose estimates (Fig 5, relative phase: R2 = .71 (N = 43); walker frequency: R2 = .23 (N = 43)). Therefore, despite the variability of recording conditions, we were able to meaningfully extract information about walking patterns from pose estimates.

Fig 5

Fig 5Comparison of relative phase and walking frequency computed from pose estimates and signals extracted from human labels of foot strike timings from 43 5-second long videos.

As an additional validity test of our method, we tested if its walking frequency estimate agrees with the typical walking cadence in the literature. We thus measured walking frequency from ankle displacement signals (Fig 5C). The average walker frequency was 1.85 Hz (SD = .32), which is very similar to the average of ~2 Hz previously reported in the literature [31]. The range of walking frequency was approximately from 1 to 2.5 Hz, appearing to reflect the cadence in naturalistic settings.

In fact, the frequency within pairs was almost identical for some walking pairs, shown by points close to the diagonal line in Fig 5D. The frequency of walkers differed by less than.1 Hz in 66% of walking pairs. This was not the case for all walking pairs, though this is understandable as the walking pairs differed (e.g., heights) and our pose estimation is inevitably plagued by noise.

We asked what the overall synchronization is in our sample as a whole. For an analysis that combined data across video segments, we quantified the mean relative walking phase for each video segment and examined its distribution across all segments.

To quantify the most prevalent relative phases, we fit a mixture of von Mises distributions model to the distribution of mean relative phase, comparing models with one to four peaks. We found that the model with three peaks provided the best fit to the data, as quantified by Bayesian Information Criterion (Fig 6A).

The means of the three components or peaks of the model were.01, -3.05, and -3.13 in radians with concentrations (normalized κ) of.17,.80, and.02 respectively (Fig 6B). The component with approximately zero mean shows that pairs walked in phase with each other. The other peak close to π show that pairs also walk in near anti-phase with each other. Note, the two peaks near π are close to each other and, accordingly, the two-peak model is comparable to the three-peak model in BIC. We have thus shown that people walking in pairs walk in phase or in anti-phase with each other.

Fig 6

Fig 6Group results on relative phase and walking frequency.

A second indicator of synchronization is a stable relative phase signal between walkers. Low variance between walkers indicates phase locking in walking patterns. Phase locking can occur at different mean phases. High variance indicates a lack of phase locking that is consistent throughout the segment, but here is difficult to separate from the effects of noise on pose estimates. We analyzed the distribution of the relative phase variance computed for each walking pair (Fig 6C).

First, we observe that the distribution is positively skewed with more walking pairs exhibiting low variance. Therefore, there is evidence of phase locking for many walking pairs in our sample. Next, we examined the variance of relative phase as a function of the mean relative phase for each walker (Fig 6D). Walking pairs with lower variance (36% of walking pairs, relative phase variance < .25) phase locked close to 0 or ±π.

The spread of the mean relative phase increases with the variance of the relative phase, as phase locking between walkers becomes less pronounced. Therefore, we find phase locking indicated by walking pairs with low variance of relative phase. In most pairs with low variance of relative phase (< .25), phase locking occurs close to 0 and ±π. Therefore, our analysis of the relative phase variance indicates phase locking in near phase or near anti-phase in 35–70% of walking pairs in our sample.

Walkers synchronize their movements using sensory feedback. Therefore, we asked if tactile feedback is crucial for synchronization to occur, by separately analyzing videos with hand contact between walkers (N = 60) and without hand contact between walkers (N = 288). We observe synchronization indicated by peaks at 0 and ±π both in cases with and without hand contact (Fig 6E and 6F). Therefore, tactile feedback is not necessary for synchronization to occur.


Source:
Tohoku University

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Questo sito utilizza Akismet per ridurre lo spam. Scopri come vengono elaborati i dati derivati dai commenti.