Sustainable Development Through Dynamic Emergency Evacuation Signage: A BIM- and VR-Based Analysis of Passenger Behavior

Abstract

To explore the influence of emergency evacuation signs on passengers’ behavior during subway fires and to enhance evacuation efficiency sustainably, this study proposes a dynamic emergency evacuation sign scheme. Utilizing a building information modeling (BIM) and virtual reality (VR) technology simulation platform, two schemes—current static signage and a novel dynamic signage system—are developed and evaluated. The research focuses on four scenarios combining varying crowd conditions (2:8 and 5:5) with signage types. Through experiments, we compare the performance of the current signage and the new dynamic signage in terms of evacuation efficiency and wayfinding difficulty. The results indicate that the dynamic identification system significantly improves evacuation efficiency, reduces incorrect route choices, and minimizes passenger confusion. Particularly in a complex scenario with a 2:8 crowd state, the dynamic signage effectively helps passengers avoid the negative impacts of group decision errors. Additionally, individual characteristics such as age, gender, spatial ability, and evacuation training experience significantly influence evacuation performance. By reducing risks, enhancing urban resilience, and optimizing evacuation processes, this study contributes to sustainable urban infrastructure safety. The findings provide a theoretical basis for designing sustainable emergency signage systems that address the social, economic, and environmental aspects of resilience in urban transportation.

1. Introduction

With the rapid advancement of global urbanization, subways have become an efficient and convenient mode of public transportation, playing a critical role in many cities. They meet the increasing travel demands of urban populations through their rapid transport capacity and effective passenger flow management [1]. However, the inherent closed nature of subway systems, combined with the complexity of underground environments and high passenger density, presents significant challenges for emergency evacuation. In particular, the need to evacuate passengers quickly and safely during emergencies such as fires is an urgent issue that requires resolution [2]. Research indicates that the efficiency of evacuation directly impacts passenger safety during subway fires and other disasters [3]. Therefore, establishing effective emergency evacuation strategies is of paramount importance.

In this context, emergency evacuation sign systems, which serve as vital tools for guiding passengers to make swift decisions during emergencies, are receiving increasing attention. Traditional emergency evacuation signs are primarily static indicators, typically set according to fixed evacuation routes. However, during emergencies, factors such as fire, smoke, or other hazards can alter the safest evacuation paths, rendering static signs incapable of dynamic adjustments based on real-time conditions. This can mislead passengers into entering dangerous areas, delaying evacuation times and increasing risks [4]. Existing research has highlighted that fixed signage often fails to adequately address the complexities and dynamic nature of real emergency situations [5,6,7,8]. Consequently, many scholars have begun to explore more intelligent and dynamic emergency evacuation systems to enhance evacuation efficiency and safety [9].

In recent years, dynamic emergency evacuation signage has become a popular research direction in both academia and engineering practice. Such systems utilize the real-time monitoring of changes in the emergency environment (e.g., fire location, crowd status) to dynamically adjust sign directions, providing passengers with optimal evacuation path guidance [9]. Numerous studies have demonstrated that dynamic signage can significantly improve evacuation efficiency and reduce erroneous decision-making [10,11]. For instance, research by Filippidis et al. validated the effectiveness of dynamic signage in enhancing evacuation efficiency in complex environments through fire simulation scenarios [11]. However, most existing studies focus on system design and simulation validation, with relatively few studies utilizing actual scenario data, particularly concerning emergency signage systems in complex public environments like subways.

Moreover, while some research has investigated the influence of individual characteristics (e.g., age, gender) on evacuation behavior, studies addressing the impact of group behavior on evacuation path choices remain insufficient [12]. In real-world situations, passengers’ evacuation behaviors are often influenced by group decisions, especially when the majority make incorrect choices, leading individuals to follow erroneous paths and further delaying evacuation [13]. Thus, effectively guiding passengers in such complex situations is a key challenge in the design of emergency evacuation signage.

Given the typically busy operations of subway stations, opportunities for conducting non-operational activities are limited and pose potential risks [14]. Consequently, immersive virtual environments can effectively simulate real conditions, reflecting passengers’ actual responses during emergencies. An increasing number of studies have confirmed the validity of VR experiments [15,16]. VR technology not only allows for the development of diverse scenarios to enrich experiments and yield various conclusions but also enables the simulation of significant disaster situations. Therefore, VR technology provides a means to replicate urgent scenarios, offering strong support for obtaining objective data on passenger evacuation behavior.

In summary, to address the aforementioned issues, this study proposes a new dynamic emergency evacuation signage system and utilizes a BIM+VR virtual experiment platform to simulate subway fire evacuation scenarios, systematically investigating the impact of the signage system on passenger evacuation behavior. The research not only compares the differences in evacuation efficacy between the existing static signage and the new dynamic signage but also further analyzes how individual passenger characteristics (such as age, gender, and spatial ability) and crowd conditions (such as group erroneous decisions) affect evacuation behavior. Through this study, we aim to provide data support and theoretical foundations for the design of emergency evacuation signage as well as guidance for optimizing emergency responses in complex public scenarios such as subways.

2. Methods

2.1. Experimental Platform

Firstly, in accordance with the technical and functional requirements for developing a three-dimensional architectural model of Hujialou Station, we conducted a thorough investigation of the actual conditions of the station. We obtained the blueprints of the station’s main building structure (as shown in Figure 1) as well as the dimensions, quantities, and placement intervals of the facilities and equipment within the station, such as the width of the ticket gates, the height of the escalators, and the length of the passages. Utilizing Revit 2018 software, we accurately replicated the main building structure of Hujialou Metro Station, along with the corresponding facilities and equipment, to create a complete architectural structure. Additionally, we applied preliminary material and color rendering to the constructed model in pursuit of a realistic effect. Next, Unity 3D software (version 2022.3.8f1c1) was used to render the station scene and fire scenario, incorporating virtual characters and developing the emergency evacuation signage plan. Finally, VR devices were employed to connect the simulated scene with the virtual environment, recreating the real situation and establishing a testing platform for the emergency evacuation signage system based on Unity and VR, as shown in Figure 2. Notably, this experiment utilized the Meta Quest Pro headset (manufactured by Meta Platforms, Inc., Menlo Park, California, USA). Meta Quest Pro is equipped with high-resolution sensors, a high-definition LCD display, and eye-tracking and facial-tracking features, providing a full-color mixed reality experience. This device has 12 GB of RAM and 256 GB of storage and includes 10 high-resolution sensors (five internal and five external), significantly enhancing the immersive experience. Participants in the experiment can control the direction and movement within the virtual environment using two controllers. The virtual simulation experiment platform outputs a total of five variables (every 0.2 s), including real-time position, real-time speed, real-time viewpoint, evacuation time, and real-time joystick control (direction and force of joystick movement). Based on these variables, the pathfinding behavior characteristics of the participants were extracted for subsequent analysis of the mechanisms influencing emergency evacuation signage and design evaluation.

2.2. Scenario Design

2.2.1. Design of the Emergency Evacuation Signage System

In emergency situations, signage must effectively capture passengers’ attention and guide them through dynamic and changing environments. Consequently, efficiency and visibility are critical considerations in the design of the signage scheme [17]. To enhance efficiency, some scholars have proposed the concept of dynamic emergency-oriented signs, which utilize a two-way arrow design to convey negative or dissuasive information through alterations in the arrows’ features [9,10,11]. Additionally, research indicates that the incorporation of auxiliary information at key decision points can significantly improve the effectiveness of emergency-oriented signage, including the integration of dynamic information signs and the implementation of broadcast early warning systems [18]. In terms of visibility, studies have demonstrated that, at equivalent viewing distances, flashing signs are more readily detected by passengers than static signs [7]. Consequently, this study incorporates a flickering design into the emergency evacuation identification scheme, selecting an arrow flicker frequency of 2 Hz based on the prior literature [19].

Building on a comprehensive analysis of previous research and considering that the combined effects of design elements may further enhance the effectiveness of signage schemes, this study devised two emergency-oriented signage designs, as detailed in

Table 1. Design scheme for emergency evacuation signage systems.

Emergency Evacuation Signage System	Design Elements
	Dissuasion Format	Flashing Frequency	Auxiliary Information
			Dynamic Comprehensive Information Signage	Broadcast
Current Signage System	-	-	-	-
	Wall-type and ground-type static emergency evacuation signage
New Dynamic Signage System	Red X symbol	2 Hz	Current location and current direction of movement	Emergency notice: A fire has occurred in the subway station. Please follow the emergency evacuation signs and evacuate immediately.
	Wall-type and ground-type dynamic emergency evacuation signage (with flashing)                     Dynamic comprehensive information signage    Broadcast

. Among them, the broadcast content is as follows: “A fire has occurred in the subway station. Please evacuate immediately following the emergency guidance signs”.

2.2.2. Crowd Flow State Design

To identify the optimal emergency-oriented signage scheme under varying external conditions and enhance passenger evacuation efficiency in subway stations, this study designed four experimental scenarios: two flow conditions (2:8 and 5:5) combined with two types of emergency-oriented signage schemes (static and dynamic dissuasive signs) [13]. In the 2:8 flow state, 80% of the avatars were programmed to select the incorrect direction at a decision-making intersection, while only 20% chose the correct path. This design aims to replicate the influence of group decisions in real-life emergencies, where poor collective choices may lead individuals to follow the majority, thereby increasing overall evacuation risk. Conversely, the 5:5 flow condition provided a balanced decision-making scenario, where 50% of participants chose the correct direction and 50% chose incorrectly.

The design of the emergency evacuation signage scheme in this study emphasizes the significance of flow conditions as a key factor to simulate passengers’ behavioral responses in actual emergency situations. In the experimental setup, a virtual evacuation character was introduced, creating two distinct flow states: 2:8 and 5:5. In the 2:8 flow state, 80% of the avatars were programmed to select the incorrect direction at a decision-making intersection, while only 20% chose the correct path. This design aims to replicate the influence of group decisions in real-life emergencies, where poor collective choices may lead individuals to follow the majority, thereby increasing overall evacuation risk. Conversely, the 5:5 flow condition provided a balanced decision-making scenario, where 50% of participants chose the correct direction and 50% chose incorrectly. This configuration allows for the observation of individual decision-making in the absence of significant group pressure, enabling an assessment of the effectiveness of emergency-oriented signage systems.

2.2.3. Route Design

The experiment selected a route within the model of Beijing’s Hujialou Subway Station, with all four experimental scenarios conducted along this route. The starting point is on the platform of Line 6, and the endpoint is on the concourse of Line 10, including three decision points, as shown in Figure 3. Additionally, the placement of the current emergency evacuation signage system is shown in Figure 4, while the placement of the new emergency evacuation signage system is shown in Figure 5. The dynamic comprehensive information signs are placed at the entrance of each decision area along the route. Ultimately, four scenarios were designed by combining the two pedestrian flow states with the two signage schemes, as shown in

Table 2. Four scenarios.

	Current Signage System (Placement as Shown in Figure 3)	New Dynamic Signage System (Placement as Shown in Figure 4)
Crowd Condition 5:5	Scenario 1	Scenario 2
Crowd Condition 2:8	Scenario 3	Scenario 4

below:

2.3. Participants

In this study, subjects were recruited from university and community populations through multiple channels, including social media platforms, university campus bulletin boards, and community center notice boards. The advertisements provided detailed information about the purpose of the study, the eligibility criteria for participants, the experimental procedure and expected duration, and the compensation for participants. Candidates were required to meet three criteria: (1) normal auditory and visual function, (2) no susceptibility to 3D-induced dizziness, and (3) proficiency in operating VR controllers. After screening, 43 eligible subjects were recruited, and the final effective sample was 39, including 21 males (53.85%) and 18 females (46.15%). The age range of the subjects was 18 to 60 years, with a mean age of 30.21 ± 9.08 years. The participants were drawn from a broad spectrum of society, encompassing individuals from different professions and varying levels of education. This diverse group included students, teachers, engineers, and office workers, among others, ensuring a wide representation of the general population. This sample size is adequate to provide robust and reliable answers to the research questions [20,21].

2.4. Experimental Procedure

In this experiment, to investigate the impact of different factors on emergency evacuation behavior, we designed two routes, comprising a total of 12 experimental scenarios. Specifically, Route 1 includes four scenarios, while Route 2 encompasses eight scenarios. To prevent learning effects, we also devised six distractor scenarios. Consequently, the experiment consisted of 18 scenarios in total, which are arranged in a Latin square design order. All scenarios were completed within two working days, with a three-day interval between the two experimental sessions. The specific experimental procedures are illustrated in Figure 6. This paper focuses solely on the four scenarios in Route 1, whereas the scenarios in Route 2 will be discussed in subsequent studies.

To mitigate the learning effect among the subjects, this study designed six interference scenes, resulting in a total of 18 scenes arranged according to a Latin square order. All scenes were completed over two working days, with a three-day interval between sessions. The specific experimental steps are illustrated in Figure 6.

Before the experiment: Step 1. The subjects completed a demographic questionnaire and provided informed consent. Step 2. The experimenter introduced the virtual reality (VR) device used in the study and performed eye-tracking calibration. Step 3. The subjects entered an informal experimental scene for equipment adaptation to ensure they were familiar with the operation of the VR equipment and experienced no adverse reactions. Step 4. The experimenter read the experimental instructions to the subjects, informing them that they were preparing to leave the station and go home, and asked them to respond according to a real situation.

During the experiment: Step 5. The participants conducted the evacuation simulation experiment according to the Latin square sequence, with each scene lasting approximately 2 to 3 min. Step 6. After each scene, the subjects were asked if they needed a break. Step 7. Steps 5 and 6 were repeated to complete nine scenario tests on the first working day. Step 8. Steps 5 and 6 were repeated to complete the remaining scenario tests on the second working day.

After the experiment: Step 9. The subjects completed a post-experiment questionnaire and received compensation of CNY 160.

2.5. Data Processing

Upon the conclusion of the experiment, the fundamental pathfinding behavior data of each participant during the route task was extracted, encompassing metrics such as time, speed, and location. Subsequently, additional required indicators were derived from this foundational behavioral data. Based on these foundational behavioral data, other required indicators were further calculated.

2.6. Index Extraction

The primary function of emergency evacuation signage is to enable passengers to quickly, smoothly, and accurately locate the safety exit to complete their evacuation. In light of two key dimensions—evacuation efficiency and wayfinding difficulty—this paper selects seven indicators to construct an index evaluation system [22,23]. In the dimension of evacuation efficiency, which includes evacuation time, evacuation distance, and average speed, these indicators directly reflect the effectiveness of the signage system in helping passengers evacuate quickly and safely. In the dimension of wayfinding difficulty, which includes the number of directional identifications, number of stops, dwell time, and number of errors, these indicators reflect the level of confusion and decision-making difficulty passengers experience during the wayfinding process. This system aims to explore the differences in passenger behavior between the new version of the signage and the current status signage, as illustrated in Figure 7 below.

2.6.1. Evacuation Efficiency

“Evacuation time” refers to the duration from the start of wayfinding to the arrival at the final destination, indicating the guidance efficiency of the emergency evacuation signage system. A shorter evacuation time implies more efficient guidance by the signage system [24]. The calculation formula is as follows:

$$T=T_e-T_s$$

(1)

where $T$ represents the evacuation time, $T_e$ is the time of arrival at the final destination, and $T_s$ is the time when the evacuation begins.

“Evacuation distance” refers to the total path length traveled by the subjects from the starting point to the end point, reflecting the influence of emergency evacuation signage on subjects’ route planning. A shorter distance indicates higher evacuation efficiency. The calculation formula is as follows:

$$d_i=\sqrt{{\left(x_{i+1}-x_i\right)}^2+{\left(y_{i+1}-y_i\right)}^2+{\left(z_{i+1}-z_i\right)}^2}$$

(2)

$$D={\displaystyle \sum d_i}$$

(3)

where D represents the evacuation distance, $d_i$ is the distance at the i moment, ($x_i$, $y_i$, $z_i$) are the spatial coordinates of the subject at the i moment, and ($x_{i+1}$, $y_{i+1}$, $z_{i+1}$) are the spatial coordinates at the i + 1 moment.

“Average speed” refers to the mean speed of passengers throughout the entire evacuation process. A higher average speed indicates greater evacuation efficiency. The calculation formula is as follows:

$$\overline{v}={\displaystyle \frac{{\displaystyle \sum v_i}}{i}}$$

(4)

where $v_i$ is the speed at time i.

2.6.2. Wayfinding Process

The “number of direction identifications” refers to instances where the subject continuously rotates their field of view for more than 0.6 s (based on experimental observations, this duration effectively reflects directional recognition behavior) with a rotation angle exceeding 9° every 0.2 s (this interval is used for calculating the rotation angle). This measure reflects the participants’ confidence in the accuracy of the information provided by the signage and their level of confusion. The calculation formula is as follows:

$$R={\displaystyle \sum R_i}$$

(5)

where $R_i$ represents the i instance of a direction identification action.

“Number of stops” is recorded when the subject ceases to move the control stick for more than 0.4 s (this duration is statistically significant in identifying a clear stop based on experimental data). A higher number of stops indicates increased confusion, suggesting that the subject frequently needs to pause to determine the direction. This reflects the continuity and effectiveness of the directional information. The calculation formula is as follows:

$$S_n={\displaystyle \sum S_{n_i}}$$

(6)

where $S_{n_i}$ represents the duration of the i stop.

“Dwell time” refers to the total duration of all stops made by the subjects, reflecting the clarity of the signage guidance. The calculation formula is as follows:

$$S_t={\displaystyle \sum S_{t_i}}$$

(7)

where $S_{t_i}$ represents the duration of the i stop.

“The number of errors made” refers to the count of decision-making mistakes made by passengers who did not select the predetermined shortest path during the evacuation process. This metric reflects the accuracy of the passengers’ decision-making. The calculation formula is as follows:

$$M={\displaystyle \sum M_i}$$

(8)

where $M_i$ represents the number of errors made at the i instance.

2.7. Data Analysis Methods

A generalized estimating equation (GEE) is an extension of the generalized linear model [25], retaining its flexibility regarding the distribution and type of the dependent variable without stringent requirements. Moreover, the GEE uniquely accommodates the analysis of repeated measures and correlated observations. Given that the data in this study exhibit non-normality, a mix of categorical and continuous variables, and repeated measurements, the GEE modeling approach is employed.

To ascertain the influence of four emergency evacuation signage systems on passenger pathfinding behavior and to determine the impact of external conditions on the efficacy of the emergency evacuation signage system, this study employs generalized estimation equations (GEEs). The independent variables include the types of emergencies, the emergency evacuation signage systems, and their interaction, while the dependent variables consist of six indicators, as illustrated in

Table 3. Model construction.

Factor		Level
Signage System		New Dynamic Signage System
		Current Signage System
Crowd Condition		2:8
		5:5
Passenger attribute	Gender	Female
		Male
	Age	Age (≥50 years)
		Age (41–50 years)
		Age (31–40 years)
		Age (18–30 years)
	Subway Riding Frequency	Frequency (more than 6 times per week)
		Frequency (3–6 times per week)
		Frequency (1–2 times per week)
		Frequency (rarely rides)
	Number of Emergency Events Experienced	Experienced Events (more than 2 times)
		Experienced Events (1–2 times)
		Experienced Events (0 times)
	Number of Evacuation Drills	Evacuation Drills (more than 2 times)
		Evacuation Drills (1–2 times)
		Evacuation Drills (0 times)
	Spatial Ability Level	Spatial Ability Level (High)
		Spatial Ability Level (Medium)
		Spatial Ability Level (Low)

below.

During the modeling process, this study adopts an exchangeable correlation structure, which uses a constant correlation coefficient for all observations of the same participant. Since the goodness-of-fit measures used in ordinary linear regression are inapplicable to GEEs, this study utilizes quasi-likelihood under the independent model criterion (QIC) to identify the optimal correlation structure [26].

3. Results and Discussion

3.1. Descriptive Statistics

Under various emergency evacuation identification schemes and flow conditions, we performed descriptive statistics on the seven selected indicators. The results are presented as mean values and standard deviations in

Table 4. Descriptive statistics.

Dimension	Indicator	Signage System	Mean (Standard Deviation)		Statistical Chart
			5:5	2:8
Evacuation efficiency	Evacuation time (s)	Current Signage System	44.52 (8.63)	54.13 (17.13)
		New Dynamic Signage System	24.75 (7.76)	31.50 (7.61)
	Evacuation distance (m)	Current Signage System	78.41 (7.85)	87.90 (29.95)
		New Dynamic Signage System	75.63 (8.72)	75.56 (13.27)
	Average speed (m/s)	Current Signage System	2.89 (0.31)	2.75 (0.28)
		New Dynamic Signage System	2.97 (0.35)	3.01 (0.39)
Wayfinding process	Number of direction recognitions	Current Signage System	5.97 (3.10)	1.8 (1.47)
		New Dynamic Signage System	0.86 (1.12)	4.71 (1.78)
	Number of stops	Current Signage System	2.71 (2.42)	4 (2.91)
		New Dynamic Signage System	2.57 (2.74)	2.28 (2.06)
	Dwell time(s)	Current Signage System	1.44 (2.56)	3.59 (4.11)
		New Dynamic Signage System	1.86 (2.99)	1.43 (1.82)
	Number of errors made	Current Signage System	1.14 (0.79)	1.51 (0.93)
		New Dynamic Signage System	0.4 (0.59)	0.34 (0.71)

. The table also includes cubic graphs that depict the seven indicators across different identification groups of passengers, allowing for an examination of potential differences.

Regarding the three indices of evacuation efficiency, the group utilizing the new emergency evacuation identification system demonstrated shorter evacuation times and distances as well as faster evacuation speeds compared to the current system. Furthermore, evacuation times and distances in the 5:5 flow state were less than those in the 2:8 state.

In terms of the four indices related to wayfinding, the frequency of direction identification using the current sign system was higher in the 5:5 flow state than in the new version, whereas in the 2:8 flow state, the frequency was lower for the current system than for the new version. In both conditions, the number of stops recorded for the current sign system exceeded that of the new system. For stationary time, in the 5:5 flow state, the current system’s stationary time was slightly less than that of the new version, while in the 2:8 flow state, it was greater. Finally, the number of errors was higher in the current sign system than in the new version, with the 2:8 flow state exhibiting a slightly greater number of errors than the 5:5 flow state.

3.2. Results of the Generalized Estimation Equation

The study employed SPSS 26 to solve the generalized estimating equation model, ultimately obtaining the model effect values and parameter estimation results, as shown in

Table 5. Model effect.

Factor		Evacuation Efficiency						Wayfinding Process
		Evacuation Time (s)		Evacuation Distance (m)		Average Speed (m/s)		Number of Direction Recognitions		Number of Stops		Dwell Time (s)		Number of Errors Made
		Wald	Sig.	Wald	Sig.	Wald	Sig.	Wald	Sig.	Wald	Sig.	F	Sig.	F	Sig.
Signage System		171.07	0.000 ***	6.84	0.009 ***	7.57	0.006 ***	12.96	0.000 ***	4.31	0.038 **	4.56	0.033 **	47.38	0.000 ***
Crowd Condition		27.48	0.000 ***	2.71	0.100	1.44	0.231	0.35	0.555	3.14	0.076 *	10.26	0.001 ***	1.79	0.181
Passenger attribute	Gender	0.66	0.414	6.38	0.012 **	0.11	0.742	0.39	0.528	0.12	0.727	3.07	0.080 *	2.13	0.145
	Gender	12.16	0.007 ***	1.62	0.654	1.52	0.678	8.16	0.043 **	0.54	0.910	13.09	0.004 ***	3.40	0.334
	Subway Riding Frequency	4.79	0.187	5.30	0.151	1.57	0.665	5.31	0.151	0.15	0.986	15.79	0.001 ***	14.56	0.002 ***
	Number of Emergency Events Experienced	4.42	0.110	7.36	0.025 **	0.11	0.947	0.28	0.870	0.38	0.828	3.15	0.207	8.43	0.015 **
	Number of Evacuation Drills	5.15	0.076 *	8.67	0.013 **	0.14	0.933	2.37	0.306	0.09	0.953	3.63	0.163	13.07	0.001 ***
	Spatial Ability Level	5.25	0.072 *	0.71	0.703	2.87	0.238	5.58	0.061 *	0.12	0.944	13.11	0.001 ***	0.23	0.893
Signage System × Crowd Condition		0.75	0.386	2.75	0.097 *	3.49	0.062 *	132.71	0.000 ***	7.99	0.005 ***	8.26	0.004 ***	3.10	0.078 *

Note: *** p < 0.001, ** p < 0.05, * p < 0.1.

and

Table 6. Parameter estimation.

	Evacuation Efficiency						Wayfinding Process
	Evacuation Time (s)		Evacuation Distance (m)		Average Speed (m/s)		Number of Direction Recognitions		Number of Stops		Dwell Time (s)		Number of Errors Made
Parameter	B	Sig.	B	Sig.	B	Sig.	B	Sig.	B	Sig.	B	Sig.	B	Sig.
Intercept	51.850	0.000 ***	84.777	0.000 ***	2.475	0.000 ***	9.439	0.000 ***	4.227	0.705	1.943	0.530	0.825	0.004 ***
New Dynamic Signage System	−19.766	0.000 ***	−2.779	0.111	0.073	0.230	−5.114	0.000 ***	−0.143	0.782	0.429	0.407	−0.743	0.000 ***
Current Signage System	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Crowd Condition 2:8	9.611	0.001 ***	9.497	0.064 *	−0.142	0.006 **	−4.171	0.000 ***	1.286	0.004 ***	2.154	0.000 ***	0.371	0.050 **
Crowd Condition 5:5	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Female	1.720	0.414	−14.481	0.012 **	−0.038	0.742	−0.224	0.528	0.825	0.727	2.024	0.080 *	−0.222	0.145
Male	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Age (≥50 years)	13.987	0.007 ***	−6.614	0.503	−0.316	0.300	1.526	0.048 **	−1.733	0.776	1.450	0.428	0.181	0.655
Age (41–50 years)	4.902	0.007 ***	−14.866	0.331	−0.091	0.691	0.774	0.013 **	2.808	0.532	11.823	0.030 **	−0.477	0.113
Age (31–40 years)	6.116	0.309	−7.210	0.596	0.099	0.758	−0.567	0.654	−0.830	0.888	13.007	0.004 ***	0.256	0.467
Age (18–30 years)	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Frequency (more than 6 times per week)	−2.753	0.708	5.873	0.813	0.468	0.309	−2.869	0.088 *	−0.451	0.972	12.548	0.020 **	0.626	0.289
Frequency (3–6 times per week)	−8.203	0.137	15.271	0.291	0.383	0.216	−2.207	0.114	−1.711	0.850	15.516	0.001 ***	0.888	0.004 ***
Frequency (1–2 times per week)	−4.991	0.380	3.319	0.764	0.270	0.401	−1.000	0.442	−1.955	0.799	14.119	0.002 ***	0.501	0.137
Frequency (rarely rides)	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Experienced Events (more than 2 times)	3.040	0.366	14.751	0.394	−0.114	0.756	0.312	0.637	−3.634	0.549	3.158	0.696	0.224	0.533
Experienced Events (1–2 times)	−3.602	0.247	−27.941	0.023 **	−0.067	0.822	0.055	0.950	0.528	0.929	3.710	0.076 *	−0.863	0.006 ***
Experienced Events (0 times)	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Evacuation Drills (more than 2 times)	−4.698	0.037 **	−38.995	0.004 ***	0.128	0.712	0.499	0.300	−1.055	0.826	−1.806	0.372	−0.786	0.000 ***
Evacuation Drills (1–2 times)	−0.116	0.956	−10.700	0.195	0.008	0.944	−0.151	0.714	0.394	0.869	2.100	0.103	−0.291	0.098 *
Evacuation Drills (0 times)	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Spatial Ability Level (High)	−1.556	0.726	9.127	0.559	0.050	0.858	−1.786	0.167	−0.890	0.888	−19.656	0.000 ***	0.099	0.722
Spatial Ability Level (Medium)	−6.345	0.185	−0.903	0.910	0.257	0.366	−2.515	0.052 *	−0.118	0.986	−18.306	0.000 ***	0.010	0.966
Spatial Ability Level (Low)	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
New Dynamic Signage System × 2:8	−2.863	0.386	−9.563	0.097 *	0.185	0.062 *	8.029	0.000 ***	−1.571	0.005 ***	−2.589	0.004 ***	−0.429	0.078 *
New Dynamic Signage System × 5:5	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Current Signage System × 2:8	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Current Signage System × 5:5	0 a	-	0 a		0 a		0 a		0 a		0 a		0 a
Scale	913.240		363.800		0.718		3.880		31.259		14.805		0.626

Note: *** p < 0.001, ** p < 0.05, * p < 0.1, “0 a” referenceitem.

.

3.2.1. Model Effect Analysis

Table 4 summarizes the effects of various independent variables on the seven indicators of passenger evacuation behavior. The results for the three indicators of evacuation effectiveness demonstrate that the signage system (p < 0.001), crowd condition (p < 0.001), and age (p = 0.007) significantly influence evacuation time, while the number of evacuation drills (p = 0.076) and spatial ability level (p = 0.072) exhibit marginally significant effects on evacuation time. The signage system (p = 0.009), gender (p = 0.012), number of emergency events experienced (p = 0.025), and number of evacuation drills (p = 0.013) significantly impact evacuation distance, with the interaction between signage type and crowd condition (p = 0.097) showing a marginally significant effect on evacuation distance. The signage status (p = 0.006) significantly affects evacuation speed, while the interaction between signage type and crowd condition (p = 0.062) demonstrates marginal significance regarding evacuation speed.

For the four indicators in the wayfinding process, the signage system (p < 0.001), age (p = 0.043), and the interaction between signage type and crowd condition (p < 0.001) show significant differences in the number of directional recognitions, with spatial ability level (p = 0.061) reflecting a marginally significant influence. The signage system (p = 0.038) and the interaction between signage type and crowd condition (p = 0.005) significantly affect the number of stops made, while crowd condition (p = 0.076) shows a marginally significant effect. Significant influences on stopping time are found for the signage system (p = 0.033), crowd condition (p = 0.001), age (p = 0.004), subway riding frequency (p = 0.001), spatial ability level (p = 0.001), and the interaction between signage type and crowd condition (p = 0.004), with gender (p = 0.080) exhibiting a marginally significant effect. Finally, signage type (p < 0.001), number of emergency events experienced (p = 0.002), number of evacuation drills (p = 0.015), and spatial ability level (p = 0.001) significantly influence the number of errors made, while the interaction between signage type and crowd condition (p = 0.078) shows a marginally significant effect on error occurrence.

3.2.2. Parameter Estimation Analysis

Furthermore, this study presents the coefficients of the fixed effects solutions of the model to further quantify the differences in each indicator at different levels of the independent variables, as shown in Table 6. A detailed analysis of the indicators at various stages is provided below.

(1) Evacuation efficiency

In the analysis of evacuation effectiveness, both signage type and crowd condition show a significant impact on evacuation time, evacuation distance, and evacuation speed. Additionally, individual characteristics such as age and evacuation drill experience affect evacuation efficiency.

Regarding evacuation time, the new signage significantly reduced evacuation time by an average of 19.766 s (p < 0.001), indicating that dynamic guidance can quickly provide the optimal path and minimize delays caused by incorrect directions or backtracking behaviors. Conversely, when the crowd condition was 2:8, evacuation time significantly increased by 9.611 s (p = 0.001), reflecting the pronounced negative impact of group behavior on individual decision-making when most individuals choose the wrong path. Additionally, older participants (aged 50 and above) experienced a significant increase in evacuation time of 13.987 s (p = 0.007), suggesting that elderly individuals are slower in emergency responses. Participants with more than two evacuation drills experienced a significant reduction in evacuation time of 4.698 s (p = 0.037), demonstrating the effectiveness of drills in enhancing evacuation efficiency [27].

Regarding evacuation distance, when the crowd condition was 2:8, evacuation distance significantly increased by 9.497 m (p = 0.064, marginal significance), indicating that the incorrect choices of the majority led participants to deviate from the optimal path. Conversely, female participants exhibited a significant reduction in evacuation distance of 14.481 m (p = 0.012), suggesting that women are more likely to choose direct and efficient routes in emergencies, minimizing detours. Participants with more than two evacuation drills experienced a significant reduction in evacuation distance of 38.995 m (p = 0.004), while those who had experienced 1–2 emergency events also saw a significant decrease of 27.941 m (p = 0.023), demonstrating that passengers with drill experience and appropriate emergency exposure tend to have shorter evacuation distances and higher efficiency. Additionally, interaction effects indicate that dynamic signage can effectively reduce path deviation. Even when most individuals choose the wrong path, the interaction between signage type and crowd condition results in shorter evacuation distances for participants, helping to avoid detours and extended paths caused by incorrect decisions. This shows that dynamic signage not only provides real-time directional updates but also guides participants in avoiding the negative impacts of group errors, thereby effectively reducing evacuation distance.

Regarding evacuation speed, crowd condition also had a significant impact, with evacuation speed significantly decreasing by 0.142 m/s (p = 0.006) under the 2:8 condition. This suggests that when the majority make incorrect choices, participants’ speed is noticeably reduced, likely due to uncertainty in path selection. Additionally, interaction effects are evident in evacuation speed. Although the 2:8 crowd condition leads to a decrease in evacuation speed due to group errors, the dynamic guidance of the new signage helps participants maintain a relatively high evacuation speed even in complex environments. This indicates that the interaction effects can partially mitigate the adverse impacts of group decision-making, allowing individuals to respond more quickly in intricate evacuation situations.

(2) Wayfinding process

In the four core indicators of the wayfinding process (number of directional recognitions, stopping time, number of stops, and number of errors), multiple independent variables and interaction terms significantly affected evacuation performance. Overall trends indicate that the new signage and crowd condition are the primary factors influencing wayfinding behavior, while individual characteristics such as age, gender, and spatial ability also play important roles in specific contexts.

In terms of the number of directional recognitions, the use of the new signage significantly reduced participants’ directional recognitions by 5.114 times (p < 0.001), indicating that the dynamic signage system helped participants quickly determine the correct path, thereby reducing confusion. Under the 2:8 crowd condition, directional recognitions also significantly decreased by 4.171 times (p < 0.001), suggesting that when most individuals choose the wrong path, they are more likely to rely on group decisions, reducing the need for independent judgment. In contrast, older participants, particularly those aged 50 and above and those aged 41–50, experienced significant increases in directional recognitions of 1.526 times (p = 0.048) and 0.774 times (p = 0.013), respectively, indicating greater confusion in path selection and a tendency to repeatedly confirm directions. Moreover, participants who ride the subway more than six times a week had a significant reduction of 2.869 recognitions (p = 0.088), reflecting their familiarity with the environment and a decreased need for directional confirmation. Participants with moderate spatial ability showed a significant increase of 2.515 recognitions (p = 0.052, marginal significance), suggesting that individuals with weaker spatial cognitive skills require more time to confirm paths in complex environments. Additionally, the interaction between signage type and the 2:8 crowd condition significantly increased the number of directional recognitions by 8.029 times (p < 0.001), indicating that even with the aid of dynamic signage, participants still felt the impact of group pressure when most chose the wrong direction, necessitating more confirmations. This reflects the strong influence of group errors on individual decision-making, as participants may still feel uncertain despite correct dynamic guidance.

Regarding dwell time, crowd condition and individual characteristics jointly influenced participants’ performance. Under the 2:8 crowd condition, stopping time significantly increased by 2.154 s (p < 0.001), indicating that group errors exacerbated participants’ uncertainty, leading them to pause longer to confirm their paths. Gender also played a role, as female participants exhibited a significant increase in stopping time of 2.024 s (p = 0.080, marginal significance), suggesting that they tend to stop more frequently to confirm directions and ensure safety. Participants who use public transportation more than six times a week experienced a substantial increase in stopping time of 12.548 s (p = 0.020), possibly reflecting their familiarity with the environment, which makes them more cautious in decision-making and leads to longer pauses. Additionally, participants who had experienced one to two emergency events also showed a significant increase in stopping time of 3.710 s (p = 0.076, marginal significance), indicating that their prior experiences make them more careful in emergencies, resulting in extended stopping times. Spatial ability also significantly impacted stopping time; participants with moderate spatial ability had a significant reduction in stopping time of 18.306 s (p < 0.001), while those with high spatial ability experienced a reduction of 19.656 s (p < 0.001). This suggests that individuals with weaker cognitive skills tend to hesitate more in complex environments, resulting in more frequent stops. Furthermore, interaction effects significantly reduced stopping time by 2.589 s (p = 0.004). This indicates that although the 2:8 crowd condition increased the complexity of path selection, dynamic signage effectively shortened participants’ hesitation time, helping them make quicker decisions. The real-time adjustment feature allows participants to swiftly determine the optimal path, reducing the need for repeated stops to confirm directions.

The results regarding the number of stops indicate that under the 2:8 crowd condition, the number of stops significantly increased by 1.286 times (p = 0.004). This suggests that when most individuals in the scene choose the wrong path, participants are more likely to feel confused, leading to more frequent pauses to confirm their routes. Additionally, interaction effects demonstrated significant reductions in the number of stops by 1.571 times (p = 0.005). Although the 2:8 crowd condition results in most individuals choosing the wrong path, the new signage effectively reduces the number of stops, indicating that it can help mitigate confusion in path selection for participants.

In terms of the number of errors made, the use of the new signage significantly reduced errors by an average of 0.743 times (p < 0.001). This indicates that the new dynamic signage effectively helps participants avoid wrong paths, minimizing the backtracking and erroneous decisions caused by incorrect fixed signage. Conversely, under the 2:8 crowd condition, the number of errors significantly increased by 0.371 times (p = 0.050), suggesting that when most individuals choose the wrong direction, group pressure exacerbates the occurrence of erroneous decisions. Frequency of use also influenced the number of errors; participants who ride public transportation three to six times a week experienced a significant increase in erroneous decisions by 0.888 times (p = 0.004), indicating that individuals who frequently use public transport may make poor judgments in emergencies due to over-reliance on their environment. In contrast, experience with evacuation drills significantly reduced the number of errors, with participants who had undergone one to two drills reducing their errors by 0.291 times (p = 0.098, marginal significance), while those with more than two drills experienced a significant decrease of 0.786 times (p < 0.001), demonstrating that drills effectively enhance the accuracy of emergency decision-making. Participants who had experienced one to two emergency events also showed a significant reduction of 0.863 times (p = 0.006), indicating that emergency experience contributes to more cautious decision-making, thereby reducing errors. Furthermore, the interaction between signage type and the 2:8 crowd condition significantly reduced the number of errors by 0.429 times (p = 0.078, marginal significance). This suggests that dynamic signage can effectively assist participants in avoiding erroneous decisions even under the pressure of complex group decision-making, helping them resist blindly following group errors and improving the accuracy of path selection.

In summary, the advantages of the new signage in guiding passengers to evacuate safely and efficiently are quite evident. By dynamically adjusting directional guidance, it helps participants quickly find the optimal evacuation path in complex emergency scenarios, significantly reducing evacuation time and minimizing path deviations. Notably, under the 2:8 crowd condition, it effectively alleviates confusion in directional recognition, reduces the number of stops, and decreases erroneous decisions. Additionally, individual characteristics such as age, gender, spatial ability, and evacuation experience play significant roles in the wayfinding process, indicating that emergency evacuation strategies should be designed with greater personalization for different groups. The interaction effects between signage type and crowd condition have a crucial impact on various indicators of evacuation effectiveness. Moreover, although the negative impact of group erroneous decisions on individual behavior is pronounced under the 2:8 condition, the new signage effectively assists participants in reducing evacuation time and distance while maintaining a relatively high evacuation speed through dynamic directional adjustments. The interaction effects further highlight the advantages of the dynamic signage system in complex environments, particularly in enhancing evacuation efficiency under the pressure of majority errors.

It is noteworthy that the outcomes of this study are closely intertwined with the concept of sustainable development. The introduction of a dynamic emergency evacuation signage system has not only enhanced evacuation efficiency during emergencies such as subway fires but also contributed to sustainable development in multiple dimensions. On the one hand, the system has significantly reduced the risk of accidents, ensuring the safety of passengers and enhancing the resilience of urban transportation infrastructure. This allows the subway system to resume operations more swiftly after disasters, thereby minimizing disruptions to urban socio-economic activities. On the other hand, it has optimized the utilization of spatial resources, improved operational efficiency, and reduced long-term operational costs, thereby achieving an enhancement in economic benefits. Furthermore, this study has taken into account diverse individual characteristics, meeting the needs of various passengers, including the elderly and women, thereby reflecting social equity and inclusiveness. It also provides a reference for emergency management in other fields and contributes to the sustainable development of cities.

4. Conclusions

Based on the analysis of experimental data from subway fire evacuation scenarios, this study proposes a new dynamic emergency evacuation signage system and explores its impact on passenger evacuation behavior. Using a BlM+VR virtual experimental platform, the study compares evacuation behavior under the current signage and the new dynamic signage system, providing a comprehensive analysis from the perspectives of evacuation effectiveness and wayfinding processes. The main conclusions are as follows:

Significant improvement in evacuation effectiveness with the new dynamic signage: The research indicates that dynamic signage shows clear advantages in indicators such as evacuation time, distance, and speed. The new signage system can adjust directional guidance in real time, effectively shortening evacuation times and reducing delays caused by incorrect path choices and backtracking. Notably, under the complex scenario of a 2:8 crowd condition, the new signage system helps passengers find the correct path more quickly, effectively mitigating the negative impact of group erroneous decisions on individual behavior.

Smoother wayfinding process with the new dynamic signage: The dynamic signage also demonstrates significant advantages in core wayfinding indicators, including the number of directional recognitions, stopping time, number of stops, and number of errors. Compared to the current signage, the new signage reduces passengers’ confusion and erroneous decisions during path selection. Especially when most individuals choose the wrong path, dynamic signage assists participants in making correct judgments quickly, thereby reducing stops and repetitive confirmations of routes, ultimately enhancing evacuation efficiency.

Impact of individual characteristics on evacuation performance: The study also finds that individual characteristics such as age, gender, spatial ability, and evacuation drill experience significantly influence evacuation behavior. Older participants exhibit poorer evacuation speed and path selection, while individuals with lower spatial ability face greater challenges in complex environments, resulting in more stops and erroneous decisions. Participants with extensive evacuation drill experience show higher emergency response capabilities and significantly lower error rates, indicating that emergency training can effectively enhance passengers’ evacuation efficiency and the correctness of path selection.

In conclusion, this study has comprehensively explored the impact of dynamic emergency evacuation signage systems on passenger behavior in subway stations. The innovative use of BIM+VR technology has allowed for a realistic and immersive simulation of emergency scenarios, providing valuable insights into passenger decision-making processes. This technology not only enhances the accuracy and reliability of the experimental results but also offers a powerful tool for future research in emergency management and evacuation strategies. Moreover, by reducing risks, optimizing resource utilization, promoting social equity, and driving technological innovation, this study has made a positive contribution to achieving sustainable development goals.

Although this study effectively examined the impact of the new dynamic signage on evacuation behavior through a virtual experimental platform, several limitations must be acknowledged. Firstly, the research employed a BIM+VR virtual experimental platform to simulate subway fire evacuation scenarios. While this approach offers a reasonably realistic representation of emergency situations, there are still differences compared to actual environments. Participants’ behaviors in the virtual context may not accurately reflect their responses in real-life emergencies. Future research should incorporate a broader range of real-world scenarios or mixed reality technologies to further validate the reliability of the findings obtained from virtual experiments. For example, conducting small-scale field tests in actual subway stations could provide a more direct measure of the practical applicability of the VR results. Secondly, this study primarily focused on basic individual characteristics such as age, gender, and spatial ability. However, other factors, including psychological influences and stress responses, have not been sufficiently explored in relation to their impact on evacuation behavior [6]. These factors may significantly affect the decision-making process in actual emergency situations. Therefore, future studies should further investigate the influence of individual psychological factors and emotional states on evacuation decision-making. Finally, plans can be made for the future to collect more experimental and real-time data, particularly passenger behavior data from different types of subway stations and under various emergency conditions. This will include passenger behavior data from different time periods, varying passenger volumes, and diverse cultural backgrounds to ensure the diversity and representativeness of the data.