Evaluation of an assistance system supporting older pedestrians’ road crossing in virtual reality and in a real-world field test [original]

Frontiers in Psychology 01 frontiersin.org

Evaluation of an assistance

system supporting older

pedestrians’ road crossing in

virtual reality and in a real-world

field test

Rebecca Wiczorek * and Janna Protzak

Research group FANS, Department of Psychology and Ergonomics, Technische Universität Berlin,

Berlin, Germany

Older pedestrians are at a high risk of becoming victims of car accidents

because they tend not to pay sufficient attention to upcoming traffic. Within

our research project, an assistance system for older pedestrians has been

developed. It detects the street and communicates with the users through a

vibrotactile interface. Two evaluation studies have been carried out in order to

understand the potential benefits and drawbacks of the developed assistance

system. One study was conducted in a virtual environment (VR) with 23

participants, aged 65+. The other experiment was a field test in a real street

environment with 26 participants, aged 65+. Objective dependent variables in

both experiments were checking for traffic (operationalized via head tracking)

and stopping in front of the street (VR study), i.e., approaching time (field test).

Workload and acceptance served as subjective dependent variables. Analysis

of the VR experiment showed significantly more head rotation with the

assistance system than without it, as well as significantly more with cars than

without cars. The same was true for the frequency of stopping. No significant

difference was found concerning workload. With regard to acceptance, the

majority of participants indicated that the system was supportive and able

to reduce risks in traffic. In the field test, results for head rotation confirmed

the findings of the VR study. Analysis showed a marginally significant higher

head rotation frequency with the alarm system than without, and significantly

different patterns of checking for traffic at marked and unmarked crossings.

However, unlike in the VR study, no differences were found in approaching

time with and without the assistance system. Approaching time was slower at

marked crossings. No difference was found with regard to workload, meaning

the use of the assistance system did not increase the subjectively perceived

workload of participants. Analysis of the acceptance questionnaire showed

a positive attachment to the assistance system. However, most reported

that they did not experience any advantage from the use of the system, and

expressed no intention to buy such a system for themselves.

KEYWORDS

older pedestrians, assistance system, road crossing, virtual reality, field study

TYPE Original Research

PUBLISHED 20 December 2022

DOI 10.3389/fpsyg.2022.966096

OPEN ACCESS

EDITED BY

Evellin Cardoso,

Universidade Federal de Goiás, Brazil

REVIEWED BY

Luca Brayda,

Nextage Inc., Italy

Mihoko Niitsuma,

Chuo University,

Japan

*CORRESPONDENCE

Rebecca Wiczorek

[email protected]

SPECIALTY SECTION

This article was submitted to

Human-Media Interaction,

a section of the journal

Frontiers in Psychology

RECEIVED 10 June 2022

ACCEPTED 21 November 2022

PUBLISHED 20 December 2022

CITATION

Wiczorek R and Protzak J (2022) Evaluation

of an assistance system supporting older

pedestrians’ road crossing in virtual reality

and in a real-world field test.

Front. Psychol. 13:966096.

doi: 10.3389/fpsyg.2022.966096

open-access article distributed under the

terms of the Creative Commons Attribution

License (CC BY). The use, distribution or

reproduction in other forums is permitted,

provided the original author(s) and the

the original publication in this journal is

cited, in accordance with accepted

academic practice. No use, distribution or

reproduction is permitted which does not

comply with these terms.

Wiczorek and Protzak 10.3389/fpsyg.2022.966096

Frontiers in Psychology 02 frontiersin.org

Introduction

Older pedestrians are at a high risk of becoming victims of car

crashes, as official statistics show [Statistisches Bundesamt

(Destatis), 2020, 2021]. In 2019, around 20% of the victims of car

crashes in Germany were 60 years and older. Moreover, their risk

of dying as a result of an accident is three times higher compared

to younger victims, which is due to their higher fragility. Seventy-

eight percent of the accidents involving older pedestrians were

caused by the older pedestrians. The official statistics also indicate

an important reason for older pedestrians’ higher involvement in

such accidents: In more than half of the cases, the older pedestrians

did not pay sufficient attention to the upcoming traffic.

Prior research supports the conclusion that insufficient

attention paid to traffic is prevalent and offers some explanations.

One important reason for a lack of attention to traffic is

engagement in parallel visual tasks. Two laboratory experiments

(Zito etal., 2015; Tapiro etal., 2016) indicate higher frequencies

of checking the ground for obstacles in older compared to younger

pedestrians. Avineri etal. (2012) found a correlation between this

ground checking behavior and a fear of falling, which increases

with age (Tinetti etal., 1994; Schott, 2008). Further, Wiczorek and

Protzak (2022) show the negative impact of visual and cognitive

tasks on hazard perception in a road crossing simulation.

Another reason for insufficient attention paid to upcoming

traffic is engagement in parallel motor tasks, namely in walking.

An observation study and a photo-based questionnaire (Wiczorek

etal., 2016) suggests that both younger and older pedestrians do

not usually stop in front of a street to check for traffic. Instead,

they tend to keep walking and move their heads to check for traffic

while approaching the street. In an EEG-experiment using a dual-

task paradigm of combined real over ground walking and visual

signal detection, it was found that the number of missed visual

signals did significantly increase from standing to walking, but

only for older participants (Protzak etal., 2021).

Within the research group FANS, an assistance system has been

developed with the aim to support older pedestrians’ road crossing.

The system was developed to detect the street rather than

approaching cars because the latter is not possible yet. In order to

detect an approaching car, the sensors used for the system need a free

field. In a lot of urban street environments, this is not possible due to

obstacles such as trees, poles and, most importantly, parked cars. If,

in the future, cars are capable of car-to-car communication or, in this

case, car-to-device communication, detection of approaching cars

will bea helpful way to increase the safety of pedestrians.

However, the aim of the current system is to detect the street and

to remind the users to refrain from any parallel activities, namely, to

stop checking the floor and to stop walking. Instead, they should

focus their whole attention on the traffic. Detection of the street has

been realized through a combination of sensor fusion and machine

learning (Qureshi etal., 2018; Qureshi and Wizcorek, 2019). The

system, which is mounted to a walking frame, detects the curb stone

using a webcam in addition to an infrared-based LEDDAR sensor.

The detection rate has been optimized using CNN algorithms up to

an efficiency of more than 99%. The system was trained to detect

only the kerbstone between the pathway and the street when

approaching the street, but not the kerbstone between the street and

the pathway on the other side of the road.

Three different interfaces (auditory, thermotactile, and

vibrotactile) have been investigated in a laboratory experiment

with older participants (Wiczorek, under review).

The one that

was both efficient and had a high acceptance rate by the older

people was the vibrotactile interface. The vibrotactile interface was

realized through vibrating cuffs, worn at the upper arms. This

placement was chosen to direct users’ attention as close to the

traffic as possible when their first reaction is a shift of attention to

the application of stimulus (Bradley, 2009).

The prototype of the assistance system is shown in Figure1.

In two experiments, the assistance system was evaluated regarding

its efficiency to increase safety during road crossing as well as with

regard to the subjective workload and acceptance of the system.

The first experiment was conducted in a virtual reality (VR)

environment, and the second one was a field test.

VR evaluation study

For a long time, pedestrian simulation has been mainly video-

based and offered no or only short walking options. Since VR

technology is evolving, more sophisticated pedestrian simulators

have been developed, using head-mounted VR technologies.

These simulation environments have a higher coupling of

perception and action and allow for real walking. For a review see

Feldstein etal. (2018).

The advantage of highly immersive VR experiments compared

to video-based simulation environments is that they are much more

realistic and, thus, provide results closer to real-life behavior. VR

experiments allow for exposure of participants to traffic, without

putting them in actual danger. Furthermore, they are more controlled

than field tests. However, it has been shown that even with high

fidelity simulation, participants still do not exactly behave as in real-

world experiments (Feldstein etal., 2016). That is why wedecided to

combine both approaches, one controlled VR setting with actual

traffic and one field test in a very quiet zone with little traffic.

The aim of the VR experiment was to investigate the behavior

of older participants while road crossing with and without the

assistance system. They walked up and down a 10 m long track

and, during the walking, were presented with two different street

scenarios. For their own safety, they were equipped with a walking

frame during the whole experiment. The experiment was split into

two parts, one where the assistance system was activated, and

another one with the system turned off.

In both experiments, frequency of turning the head to the left

and the right, stopping in front of the street, i.e., approaching time,

1 Wiczorek, R. (under review). Evaluation of thermotactile and vibrotactile

cues to improve hazard perception of older pedestrians.

Wiczorek and Protzak 10.3389/fpsyg.2022.966096

Frontiers in Psychology 03 frontiersin.org

workload measures, and acceptance questions served as dependent

variables. It was expected that older pedestrians would stop more

often and move their heads more often when using the system than

when not using it. The workload measure was done to check

whether the assistance system would increase the workload as an

unintended side effect. No explicit hypotheses regarding acceptance

questions were made. They rather served to learn more about

participants’ attitudes towards the assistance system.

Materials and methods

“Ethik-Kommission des Instituts für Psychologie und

Arbeitswissenschaft (IPA) der TU Berlin” approved the study

under the name: “VR-Studie zur Wirksamkeit eines vibro-taktilen

Assistenzsystems für die Straßenquerung“(serial numbers

WI_06_20180817). All procedures were performed in accordance

with the Declaration of Helsinki, in compliance with relevant laws

and institutional guidelines. Written informed consent was

obtained from each participant and privacy rights were observed.

Participants

Twenty-three older subjects between the age of 65 and 83

(M = 73.3; SD = 5.6) were included in the analysis of this study. Ten

of them were male and 13 were female. They all walked on foot on

a regular basis. Participants were recruited via a participant tool

of the research group fans. For participation, they received a

compensation of 12€ per hour.

Research environment

The experiment took place in the “Berlin Mobile Brain/Body

Imaging Lab” (BeMoBIL) of the Dep. of Biological Psychology and

Neuroergonomics, Technische Universität Berlin. Participants

wore HTC VIVE VR glasses and were additionally equipped with

five trackers (feet, hands, and belly). The trackers and the glasses

were tracked by a room-wide installed camera system. Figure1

presents a participant with HTC VIVE during the experiment.

The scenarios were programmed with Unity. The VR scenes

covered a 10 m × 5 m corridor. Two street scenarios had been

developed for the experiment. Both showed urban environments.

Pictures of each scene are presented in Figures2, 3. For reasons of

safety and logistics, it was decided not to use a height difference

between the street and the footpath. Instead, the crossing consisted

of a so-called “drop kerb,” which in Germany is often realized by

raising the street instead of lowering the kerbstone. That allowed

participants to walk on even ground through the whole scene,

with a consistent view in the VR.

The whole walking distance inside the VR environment was

10 m, of which 7 m were in the street scene and 1.5 m to turn

around at each end. Participants started 3.5 m before the curb

stone. Vibration feedback was triggered when the subject was

2.25 m away from the kerbstone. This distance was chosen for

practical reasons to assure enough time to check for cars. When

entering the street, participants walked 3.5 m until the scene

stopped automatically (0.5 m before the end of the virtual street).

In half of the scenes, cars were crossing. They crossed before

and/or after the vibration feedback. Cars drove with a velocity of

28 km per hour. Cars appeared in a pseudorandom order. The time

the cars started was varied to make prediction impossible for

participants. Cars were triggered by the distance of the subject to

the street. This distance varied between 5.5 m, 4 m, and 3 m before

the vibration feedback, and 2 m, 1.5 m, and 1 m after the vibration

feedback was given. The blocks consisted of 18 trials each, with

nine trails containing cars. As it was a double lane road, cars could

come from both directions. The number of cars from the left and

right was counterbalanced.

FIGURE1

A participant wearing the HTC VIVE and using the walking frame during the VR experiment.

Wiczorek and Protzak 10.3389/fpsyg.2022.966096

Frontiers in Psychology 04 frontiersin.org

When a new scene started, participants could decide when to

start walking. When participants arrived at the other side of the

road (i.e., 0.5 m before the end of the road), the scene ended,

participants went into a grey room, where they received text-based

information in addition to a symbol that indicated to turn around.

They were then instructed to place their feet at a marked position

on the floor. When they were in the right spot, the next scene

started. The two different street environments were alternating.

The subject’s body was represented by either a male or female

avatar to improve immersion (Slater, 2009). The representation of

the walking frame, followed the hybrid prototyping approach

(Exner etal., 2016). It was physically present and touched by the

subjects, as well as also equipped with a tracker and visually

represented in the VR.

The original assistance system that detects the kerbstone was

simulated in the VR experiment. Thus, unlike the real system the

one used in the VR experiment was 100% reliable.

Procedure

At arrival, participants filled in the consent form and read the

instructions. Afterwards, they conducted the MoCA (Montreal

Cognitive Assessment, Nasreddine et al., 2005), an acuity test

(Landolt ring chart), and a test regarding contrast sensitivity (Pelli-

Robson chart), before answering a simulator sickness

questionnaire. Then, participants read the VR instruction, and

trackers were put on the hands, feet, and belly. The avatar was

calibrated to the person’s height. When everything was prepared,

participants had a 10- to 20-min training phase to familiarize

FIGURE2

Street scene 1 of the VR experiment.

FIGURE3

Street scene 2 of the VR experiment.

Wiczorek and Protzak 10.3389/fpsyg.2022.966096

Frontiers in Psychology 05 frontiersin.org

themselves with the VR environment. Before and after the training

phase, they answered the SSQ (Simulator Sickness Questionnaire,

Kennedy et al., 1993). Afterwards, the assistance system was

introduced and its functions were demonstrated. Participants were

instructed to cross the streets as normally as possible, i.e., to take

safe decisions, but not to beunnaturally cautious. They were

informed that the assistance system was there to support road

crossing. However, they did not receive any instruction on how to

behave as a response to the vibration signal. It was not mentioned

that the system should support stopping and checking for traffic.

The actual experiment consisted of two blocks with 18 trials each.

Both blocks contained nine trials with cars and nine trials without.

The walking frame was used during the entire experiment for

safety reasons, but one block was conducted with the assistance

system switched on, and the other one with the system switched

off. The order of blocks was counterbalanced. After each block,

participants filled in the NASA TLX (NASA Task Load Index, Hart

and Staveland, 1988). When the experimental blocks were over,

they answered the acceptance questionnaire. Finally, participants

received financial compensation and were thanked for

their participation.

Dependent measures

Objective dependent measures were stopping (both feet on

the floor with a max. length of 5 cm between feet, for a min. time

of 2 s) frequency per block and head rotation frequency (straight,

medium left, medium right, complete left, and complete right).

Every orientation was defined as a window of 36° in the rotation

field of 180° in front of the participant. NASA TLX served as a

measure for subjective workload. Acceptance was assessed via the

three questions that are listed in Table1.

Results

Stopping frequency and workload were analyzed with 2 × 2

ANOVAs with repeated measures. Head rotation frequency was

analyzed with a 2 × 2 × 5 ANOVA. Significance level alpha was set

to 0.05. Values between 0.05 and 0.1 are classified as marginally

significant. Acceptance was analyzed descriptively. Assumptions

of sphericity were tested using the Mauchly test. In case of

violation, Greenhouse–Geisser corrected values are reported.

Stopping frequency

Stopping frequencies were analyzed using the sum of all stops

for the respective number of trials (i.e., 18 trials with/without cars,

and 18 trials with/without an assistance system). The main effect

for cars revealed significance with a large effect size, F(1, 22) = 9.27;

p = 0.006; η2p = 0.3. When cars were crossing, participants stopped

with a higher frequency (M = 2.72 SD = 3.82) than without cars

(M = 1.09; SD = 2.67), but the standard deviation was higher with

cars than without. Analysis of the main effect of the assistance

system revealed only a marginally significant result but had a large

effect size, F(1, 22) = 3.53; p = 0.07; η2p = 0.14. Participants stopped

more often with the assistance system (M = 2.2, SD = 3.38) than

without it (M = 1.63, SD = 3.1), and the standard deviation was

similar for both conditions. The interaction effect did not reveal

significance. Results are presented in Figure4.

Head rotation frequency

Head rotation frequencies were analyzed for the five

orientations within single trials. The main effect of cars was

significant and based on a large effect size, F(1, 22) = 7.68; p = 0.01;

η2p = 0.26. Participants moved their heads more frequently when

cars were passing (sum of all orientations: M = 8.0, SD = 6.1) than

without cars (sum of all orientations: M = 6.4, SD = 4.16), but the

standard deviation was higher with cars. The main effect of the

assistance system revealed significance and the effect size was large,

F(1, 22) = 5.72; p = 0.03; η2p = 0.21. The frequency of head movement

was higher with the assistance system (sum of all orientations:

M = 7.92, SD = 5.14) than without the system (sum of all orientations:

M = 6.51, SD = 5.07), and the standard deviation was similar for both

conditions. Results are presented in Figure5. The main effect for

orientation was also significant with a very large effect size, F(1.76,

38.81) = 44.78; p < 0.001; η2p = 0.67. The highest frequency was found

for the straight head orientation (M = 2.12, SD = 1.16), followed by

the medium right (M = 1.62, SD = 1.17) and the medium left

orientation (M = 1.67, SD = 1.38), and the lowest frequencies were

found for complete right (M = 0.91, SD = 0.66) and complete left

orientations (M = 0.89, SD = 0.74). The interaction between the

assistance system and orientation was marginally significant, with a

medium effect size F(2.39,52.52) = 2.82; p = 0.06; η

= 0.11. The other

interactions were not significant. Results are presented in Figure6.

Workload

The overall workload did not differ significantly. It was

perceived as low (on a scale of 0–100) in both conditions, with the

TABLE1 Frequencies and percentages of answers to the three

acceptance questions regarding the assistance system in the VR

study.

Questions Answers

The

assistance

system

increases

traffic safety

Totally

disagree

Rather

disagree

Indifferent Rather

agree

Totally

agree

1 4.3% 2 8.7% 4 17.4% 9 23.1% 7 30.4%

A lot of

people would

like the

assistance

system

Totally

disagree

Rather

disagree

Indifferent Rather

agree

Totally

agree

0 0% 5 21.7% 5 21.7% 11 47.8% 2 8.7%

Would

youbuy such

a system?

Most

unlikely

Rather

unlikely

Indifferent Rather

likely

Most

likely

4 17.4% 2 8.7% 10 43.5% 3 13% 4 17.4%

Loading more pages...