Pedestrian Crowd Management Experiments: A Data Guidance Paper

The experiments were performed in a circuit training model. This means that three experimental setups were performed at the same time at three different sites, and participants were guided from one site to the other in designated groups. The three groups were marked with wristband colors: red, green or blue. The experimental sites were labeled alphabetically ‘B’, ‘C’ and ‘D’ (Fig. 1). The experimental sites were separated by black curtains that shielded the view but were not sound proof. To limit the view on to the experiment, the waiting areas within the experimental sites were shielded by curtains as well. Each day (day 1-3) consisted of six experimental time slots lasting 1 hour each, therefore participants attended each experimental site twice a day, but never participated twice in the same experimental setup, as those changed from one time slot to the next. A rough time schedule is shown in Fig. 2.

The following experiments were performed in site B:

• •

  Train Platform Experiment (day 1-3; \srefsubsec:TrainPlatformExperiments)

The following experiments were performed in site C:

• •

  Crowd Management Experiment (day 1-3; \srefsubsec:CrowdManagementExperiments)

• •

  Single-File Experiment (day 4; \srefsubsec:OvalExperiments)

• •

  Personal Space Experiment (day 4; \srefsubsec:PersonalSpaceExperiments)

The following experiments were performed in site D:

• •

  Boarding and Alighting Experiment (day 1-3; \srefsubsec:BoardingAndAlightingExperiments)

• •

  Tiny Box Experiment (day 1-3; \srefsubsec:TinyHouseExperiments)

• •

  Bottleneck Experiment (day 4; \srefsubsec:BottleneckExperiments)

Day 4 was different in that participants were divided into only two groups: Group yellow consisted of 80 people and group red of 120 people. Group red took part in the experiments at site D in all six time slots. Group yellow took part in the experiments at site C for the time slots 1 to 3 and also came to site D for time slots 4 to 6.

After submitting the registration form, the potential participants received an e-mail that their registration had been received. People were assigned to days based on their statements of availability and evenly divided among the days if they were available for multiple days. People were only able to register for one of the three first days and additionally for the fourth day.
After allocation to the days, participants received an e-mail with allocation information. Two weeks prior to the experiments, a reminder was sent including current information on the hygiene and safety concept. The hygiene concept to protect against infection by Covid-19 was a necessary requirement of the authorities and institutions involved. One week prior to the experiments, participants received an e-mail with a reminder of their personal assigned dates and important things to remember and bring along with them (e.g. ID card, comfortable shoes, wear dark clothes).

The wristbands had three different colors (red, blue, green) and were handed out alternately. That way participants were divided into three experimental groups on arrival. Participants who arrived in social groups were therefore split among our experimental groups, although it cannot be completely ruled out that people who know each other were in the same experimental group. The wristbands were labeled with numbers (Fig. 3 a) referring to the number associated with the personal ID code. The number was used for all questionnaires throughout the course of the day, to allocate sensor information and trajectories to a participant without revealing personal information.

After registration participants entered a course that led them to a sequence of stations. At these stations, information was collected and subjects were provided with markers and utensils:

• •

  measuring height

• •

  measuring shoulder width (\freffig:registration b)

• •

  measuring weight

• •

  checking color of top and handing out black long sleeve shirt if necessary

• •

  applying shoulder markers to top (\freffig:registration c)

• •

  putting on green hat and attaching personal code (\freffig:registration c, d)

• •

  checking correct fit of hat with code and telling people to leave the hat on for the entire day

• •

  time for questions

• •

  final check if declaration of informed consent was signed and questionnaires were filled out correctly

• •

  targeted addressing of suitable people to ask if they were willing to wear additional sensors (3D motion capturing suit, heart rate sensor)

After completing the measurement course participants could check their bags at a cloakroom and proceed to a large waiting area.

In the first year of the Covid-19 pandemic, Germany’s regulations prohibited people from getting together in large groups and required them to keep a distance of 1.5 m between people not belonging to the same household. During the summer of 2021 these restrictions were dropped for vaccinated people. In public it could be observed that people kept wearing masks and kept their distance to other people to a large extent on a voluntary basis.

Because our experiments were designed for situations in which high densities may occur, we attempted to mitigate the behavioural changes described above. In order to get people accustomed to larger groups and higher densities again, we performed an ‘icebreaker’ experiment. Participants were not informed about the icebreaker experiment, which was performed as part of the walkway to the first experiment. After registration people waited in a large waiting area (\freffig:sites green shading). When the registration was finished up to 100 people were asked to walk to the first experiment at a time (based on the color of their wristband). A person in charge, responsible for an experimental area, walked them into a corridor with two doors (\freffig:sites blue shading). The person in charge asked people to wait until the last participant had arrived in the corridor. The rear door was closed when everyone had arrived (the density in the room was about 1 P/m²). Then the person in charge waited for another few minutes before releasing participants into another open space. The icebreaker was performed once every morning for each group.

To assess the extent to which participants were influenced in their actions by the thought of the pandemic, everyone was given a questionnaire about perceptions of various risks (focusing, in particular, on perceptions of risks of Covid-19 infection) and about the potential influence of the Covid pandemic on the experiments, at the end of each experimental day. The questions were answered on a 7-point scale from “strongly disagree” (1) to “strongly agree” (7). The questionnaire started with the general items about whether participants felt uncomfortable in the crowds during the experiments. Participants were then able to rate how much the following seven factors influenced their discomfort: Crowding, concern about contracting Covid, concern about contracting an other illness, unclear instructions, physical exertion, stress caused by the experimenters, being with many people. Two more questions directly addressed the mental engagement with Covid. In addition, participants could indicate in which setting (e.g., in the registration course) and in which type of experiment (e.g. bottleneck) they were most concerned about Covid with a yes/no answering format. In the two final questions, subjects estimated whether they would have behaved differently before the pandemic and indicated whether they had already been in a crowd before the experiments since the onset of the pandemic. N = 1000 participants filled out the questionnaire on the four experimental days. Descriptive statistics are reported for the questionnaire data (\treftab:covidStatistics).

The results of the questionnaire suggest that the Covid pandemic did not have a major impact on the answers of the participants. The concern about infection was reported to be low (mean value M between 1.98 and 2.45). Subjects self-reported that their actions were not significantly different than before the pandemic (M = 2.69). A self-selection effect certainly plays an important role for these results: Presumably only people who were not very concerned about Covid signed up for the experiments. This can also be seen in the last question. The statement that they have already been in a crowd elsewhere was often agreed with (M = 4.02). Furthermore, these results reflect the extensive safety measures that apparently reduced the fear of contagion. In general, a low expression of discomfort was reported in the experiments (M = 2.61). The factors that caused the most discomfort were crowding (M = 3.37) and physical exertion (M = 3.49). Subjects thought most strongly of Covid in the morning registration and measurement course (38 % answered with yes). We explain this by the fact that the registration came immediately after a mandatory Covid test. This meant that the topic of the pandemic was very present at that moment. On days 1-3 participants thought about Covid most often in experiment site C (26.2 %), followed by site D (16.1 %) and B (10.4 %). On day 4 participants thought about Covid quite a lot in the experiment site D (37.9 %) and very little in experiment site C (6.6 %).

The outer dimensions of the platform were 7 m x 20 m x 0.8 m. The ascent and descent was realized by stairs secured with railings and organized in a way that resembled one primary access and three stairs to “board the train” that only a few people could stand on at the same time. The stairs at the narrow side were 3 m wide while the stairs at the long side of the platform were each 1.5 m wide. The smaller stairs were movable and the positions for save attachment to the platform were only visible for helpers moving the stairs. The platform’s edge was marked with adhesive tape (width 0.05 m). White adhesive tape marks the safety distance (0.8 m) from the edge of the platform, which is a standard for platforms in Germany. A loudspeaker box with recordings of railroad sounds was placed under the platform during all of the experiments to reduce the influence of sound from the neighbouring experimental sites.

Cameras were mounted to record the experiment. They are listed in Table 2. Experimental runs in which 3D motion capturing data were recorded are listed in Table 11. The mood of the participants (cf. \srefsubsec:moodbutton) was recorded for all runs. Trajectories were generated as described in Section 4.2. The coordinate origin’s location was centred in relation to and 0.5 m from the short platform border (Fig. 6a), with the y-axis pointing to the left, parallel to the longer edge and aligned with the midpoint of the large stairs. The data of the Train Platform Experiments are provided online [7].

Instructions according to the motivation and the run were given directly in front of the entrance to the experiment before participants could see the experimental setup. At the time of the instructions, participants were separated from the experiment by a black curtain. If a norm specification was given, every participant was given a slip of paper with a note on it saying either “In the following experiment, behave as you always would” or “Imagine that you are a very selfish person. Push yourself to the front during the experiment”. The percentage refers to the amount of paper notes prescribing normal behaviour. After each run, participants were directed back to the area where the instructions were given and questionnaires distributed. HRV Sensors were collected (if given) after each group had finished their respective runs. Within the framework of the experiments, a comparison of the estimated and the physically measured density of people was carried out. Some assessors had been previously trained to the Level of Service and had been given further knowledge of density estimation, while others were untrained. Time-dependent densities ranging from low to medium to high were estimated [8] and documented. In addition, positive and negative factors influencing the density estimation of the observers were surveyed using questionnaires.

The outer dimensions of the experimental area were 7 m x 20 m. The line-up gates construction was 2.5 m x 3.3 m x 1.18 m (length x width x height) with a passage width of 0.5 m each. The line-up gate construction at the 90°bend setup was 30 cm wider. Police barriers with dimensions of 2.0 m x 0.94 m x 1.1 m were used to set up the structure grid.

Cameras were mounted to record the experiment and are listed in Table 3. Experimental runs in which 3D motion capturing data were recorded are listed in Table 11 and runs in which HRV data were recorded in Table 10. The mood of the participants (cf.\srefsubsec:moodbutton) was recorded for all runs. Trajectories were generated as described in Section 4.2. The coordinate origin was located where participants enter the experimental area, with the y-axis pointing in walking direction and aligned with the midpoint of the middle entry gate (\freffig:setup_siteCd1-3_scetch). The data of the Crowd Management Experiments are provided online [9].

Instructions to start or to stop were given by a person standing between the two experimental setups without using technical amplification. Participants walked in the oval for at least 2 minutes or until they had walked one round at very high densities. The width of the ovals walking paths was 0.8 m with a circumference of 14.97 m as measured from the middle of the indicated walking width (Fig. 9a). The course was indicated by colored markers on the floor. Two oval experiments were performed at the same time. They were separated by a wooden wall (Fig. 9b).

Cameras were mounted to record the experiment and are listed in Table 4. Experimental runs in which EDA and HRV sensors were recorded are listed in Tables 9 and 10. Trajectories were generated as described in Section 4.2. The coordinate origin was located at the lower side of the two ovals’ in the axis of the screen wall (Fig. 9a). The data of the Single-File Experiments are provided online [10].

Instructions were given without using technical amplification as the groups were small. Eight runs of four minutes each were performed in total (a list of performed runs can be found in Appendix A.6).

The dimensions of the experimental area were 12.1 m x 5.3 m. The experiment was performed next to the Oval Experiments outlined in the previous subsection. There was no visual shielding between the two experiments and it was possible for participants to pass unhindered between the two experimental sites. Participants with EDA and HRV sensors were placed at positions indicated as red dots in Figure 10a. Passing participants were participants who were not currently on runs in the neighbouring Oval Experiment. Questionnaires were completed at the end of the whole experiment set including the runs of the Oval Experiment.

Cameras were mounted to record the experiment and are listed in Table 5. Experimental runs in which EDA and HRV sensors were recorded are listed in Tables 9 and 10. No trajectories were exported for this experiment. The data of the Personal Space Experiments is provided online [11].

Instructions indicating ‘the arrival’ and ‘the departure’ of the train as well as the opening and the closing of the doors were given without using technical amplification from a person standing behind the waiting/boarding pedestrians. Special announcements were made via slips of paper or by directly addressing individuals by the investigators to make individual, targeted announcements when necessary to achieve the study objective. These kinds of announcements as well as handing out luggage were done in the waiting area to the left of the experimental area. Persons assigned to a group got sticky dots of the same color and had the instruction to stay together during the boarding process. For the variation of norm, the percentage in the list above refers to the amount of paper notes prescribing ‘normal’, considerate behaviour whereas the remaining persons got the information that pushing is allowed. Where questionnaires were completed, this was done after the respective runs in the waiting area.

The outer dimensions of the experimental area were 20 m x 20 m. The inner dimensions of the train car were approximately 9.2 m x 3 m (the exact dimensions can be extracted from Fig. 11a) and aim to mimic a typical local train in Germany with the measurements $w_{door}=1.2\,m$, $w_{const1}=0.5\,m$, $w_{const2}=0.8\,m$, $w_{const3}=4.0\,m$, $w_{aisle1}=0.9\,m$, $w_{aisle2}=2.2\,m$ as indicated in Figure 11a.

Cameras were mounted to record the experiment and are listed in Table 6. Experimental runs in which 3D motion capturing data were recorded are listed in Table 11. The mood of the participants (c.f.\srefsubsec:moodbutton) was recorded for all runs. Trajectories were generated as described in Section 4.2. The coordinate origin was located on the axis of the front of the bottleneck in the middle between the two bottleneck sides. The data of the Boarding and Alighting Experiments are provided online [12].

The tiny boxes were four wooden boxes of 1 m $\times$ 1 m and a height of 1.5 m (Fig. 12). Participants could enter and exit the boxes through a one-sided swinging door. The experiments were performed in a delivery channel close to experimental site D (cf. Fig.1) to shield participants as well as possible from acoustic and visual influence of the experiments performed at the same time. Participants were chosen based on age and gender from the respective groups taking part in the experiments at site D. Instructions were given without technical amplification as the groups were small. Where questionnaires were completed, this was done after the respective runs in an area in front of the delivery channel.

Cameras were mounted to record the experiment and are listed in Table 7. Experimental runs in which EDA and HRV sensors were recorded are listed in Tables 9 and 10. No trajectories were exported for this experiment. The data of the Tiny Box Experiments are provided online [13].

The experiments were performed on day 4 at experimental site D. The announcer and further observers were standing on a scissor lift that was parked behind the bottleneck construction and raised to a height of several meters to have a good overview. All announcements were made with a microphone connected to a portable loudspeaker. To increase the initial density, the participants in the first row were asked to stay in place while everyone else was asked to take one step forward, before each run. The intended initial density was $1\,P/m^{2}$. Whenever special targeted announcements were necessary to achieve the study objective, they were made via slips of paper or by the investigators directly addressing individuals. In ‘normal’ condition runs participants were instructed as follows: “You are in a crowd where people walk through a door at a normal pace. You yourself move purposefully, but without haste.” In ‘hurry’ conditions the instructions were “You are in a crowd where people are in a hurry to pass a door. You yourself are also moving briskly”, and in ‘full commitment’ conditions “You are in a crowd where everyone wants to pass through a door as quickly as possible and pushes their way through. You yourself do everything you can to get to the front and through quickly as well.”

Figure 13 shows snapshots of two example experiments and sketches of the bottleneck construction. The outer dimensions of the experimental area were 20 m by 20 m. The bottleneck construction consisted of an 4 m x 2 m x 1 m aluminium frame with gray plastic panels, weighing 250 kg per side. They were each visually extended to be 6 m long by adding trade fair walls. Each side was secured against slipping with anti-slip mats and 750 kg concrete blocks which were bolted to the bottleneck construction. Participants were to maintain their motivation until they crossed a finish line 8 m behind the bottleneck. The way to the finish line was marked with barrier tape. Beyond the finish line, participants could return to the line-up area by turning to either sides. Where questionnaires were to be completed, this was done after the respective runs in an area to the left of the experimental area.

To take safety precautions, a practised crowd manager was present at the experiments equipped with an air pressure horn. The horn was activated whenever a participant indicated discomfort during an experiment by calling out ‘stop’ aloud or if the crowd manager himself identified a situation as critical or potentially harmful. At the beginning of the day, all participants were trained in what was to be done in the case of the horn being activated. The procedure included immediately stopping in the current position without further movement. Designated helpers that were close to the crowd at all times started tapping people at the shoulder once the crowd had come to a full stop. On the signal of ’shoulder tapping’ participants were allowed to turn around and move to the far back of the experimental site. The procedure was continued until every person was tapped at the shoulder. Apart from the test run where the horn was activated on purpose, two runs (4D250, 4D280) were aborted and resolved in the way described above.

Cameras were mounted to record the experiment and are listed in Table 8. Experimental runs in which 3D motion capturing and pressure sensor data were recorded are listed in Table 11. Pressure sensor data (cf. \srefsubsec:Pressure) and the mood of the participants (cf. \srefsubsec:moodbutton) were recorded for all runs. Trajectories were generated as described in Section 4.2. The coordinate origin was located on the axis of the front of the bottleneck walls in the middle between the two bottleneck sides. The data of the Bottleneck Experiments are provided online [14]. The scientific content of some of the bottleneck experiments in this series is part of the CrowdDNA project.

In total, 21 cameras were mounted in order to perform the above tasks. Cameras intended for extracting trajectories and serving documentation purposes were mounted under the ceiling ($\approx$ 8.65 m) facing straight downwards. Camera views were overlapping and mounted in such a way that the occlusion of people was minimized. All cameras used for trajectory extraction were backed up since image loss would have been fatal for the experiments. The approximate fields of view at head height for cameras mounted in the main experimental sites are shown in Figure 14 and listed in Tables 2-8. The camera types and settings are listed in Table 1.

The sensor was attached to the non-dominant hand of the subjects using a wristband. There were two cables attached to the wristband. These cables connected the two measurement electrodes to the sensor. The electrodes were structural non-woven electrodes with special gel/solid gel and a diameter of 55 mm, which was cut to size if necessary. The electrodes were glued to the palm of the hand below the little finger so that the gel surfaces of the electrodes did not overlap. If the electrodes did not hold well, they were fixed with leukotape. The EDA sensors were always attached by the experimenters and worn for a maximum of one hour. Between different subjects, the sensors were not read. The separation of the data was done in the aftermath by cutting up the individual experiment blocks. The sensor number and the subject number for the day were noted and thus the data of the sensor and the remaining data of the subjects could be linked. EDA data was recorded in runs listed in Table 9.

The sensor was placed below the chest using disposable electrodes. The electrodes contained a highly conductive wet gel and a high quality Ag/AgCl sensor. They had a decentred connection to reduce motion artifacts. The ECG sensors were frequently attached by the subjects themselves. Between different subjects, the sensors were not read. The separation of the data was performed by cutting up the individual experimental blocks. The sensor number and subject number for the day were noted and thus the sensor data and the rest of the subject data could be linked. Heart rate data was recorded in runs listed in Table 10.

On either side of the bottleneck, a pressure sensor was attached vertically at a height of 0.97 m for the lower edge (\freffig:pressure_bottleneck a). The short side of the sensor was bent around the corner to place 10 cm of the measurement area inside the bottleneck and 47.8 cm in front of it (\freffig:pressure_bottleneck b). Teflon foil was spread over the pressure sensors to reduce shear forces and ensure a secure attachment. Each sensor was connected to a laptop recording pressure with the I-Scan software using a sampling rate of 60 fps.

Furthermore, two participants were equipped with flexible pressure sensors on their body, each with two upper arm sensors and one sensor for the back. Xsensor LX210:50.50.05 [27] has 2500 measuring cells providing pressure measurement on an area of 25.4 cm $\times$ 25.4 cm on the participant’s back. The arm sensor (Xsensor LX210:25.50.05 [28]) covers an area of 12.7 cm $\times$ 25.4 cm with 1250 measuring cells. All sensors were calibrated in advance by the manufacturer resulting in a pressure range of 0.14 Ncm^-2 – 10.3 Ncm^-2.

For easy wearing, the three pressure sensors were tucked into designated pockets of a specific T-shirt (Fig. 16) and connected to a tablet. The tablet, which was carried in the chest pocket throughout the experiments, used the Software Xsensor Pro V8 to capture pressure at a sampling rate of 25 fps. In order to receive as much pressure as possible at the central part of the back, the volunteers who wore the shirts were 1.91 m and 2.04 m tall. Unfortunately, no pressure data from the Xsensor sensors in the T-shirts were recorded during the experiments.

We used the 3D motion capturing (MoCap) system MVN Link by Xsens to track the full body motion of a person in the crowd [29]. While optical MoCap Systems need a free line of sight between the tracking points on the body and a set of cameras, the Xsens MoCap system uses inertial measurement units (IMU) as sensors. These IMUs measure the acceleration, the angular rate and the magnetic field strength and a line of sight between the body and a camera is not necessary. Therefore, it is possible to capture the full body motion even in dense crowds.

Each MVN Link suit (\freffig:xsens1) is equipped with 17 IMU sensors on predefined independently moving body segments. The measurement can be triggered manually and the recorded data is stored locally on a body-pack in the suit. Thus, the measurement is self-contained and the data can be downloaded afterwards.
After doing a calibration procedure and taking detailed body dimension measurements, the MVN Analyze software then calculates the full body motion based on a biomechanical model from the measured data set. The processed data includes the orientation, position, velocity, acceleration, angular velocity and angular acceleration of each body segment as well as the angles of joints and the location of the centre of mass. The data can be exported either as xml file or as biomechanical c3d file.

Because the IMU based motion capturing is self-sufficient and based on relative measurements only, the absolute positioning in space suffers from a drift which can accumulate over time. The head trajectories extracted from camera recordings, however, have a small positioning error. Therefore, we used a hybrid tracking algorithm [31] to combine both data sets. In particular, that means that the position of the biomechanical model was shifted and rotated to match the head position and orientation of the camera trajectories.

On all days, we equipped 20 people with an Xsens MVN Link Motion Capturing system. On experiment days 1-3, these persons were part of the red group (Fig. 2), and on day 4 they took part on experiment site D, namely the bottleneck experiment. 3D motion capturing data were therefore recorded in runs listed in Table 11.

The terminals were attached to man-rails and placed so that participants passed them after each run. Care was taken to ensure that the grids were positioned in such a way that the walking path was affected as little as possible and no backlog was created. Participants were actively asked to press a button after each run. In the Train Platform Experiments, the terminal was placed in the corridor leading participants from the area where they filled out questionnaires back to the waiting area in front of the experiment. In the Crowd Management Experiments, the terminal was positioned 15 m after passing the entry gate on the way back to the line-up area. In the Boarding and Alighting Experiments, the terminal was placed next to the waiting area (behind the train car for runs marked as ‘reversedirection’). In the Bottleneck Experiments, one terminal was placed at each side of the bottleneck. Participants passed the terminal on their way back to the line-up area, regardless of whether they turned right or left after passing the finish line. No mood buttons were placed in the Tiny House, Oval or Personal Space Experiments.