Crowd Motion Research Articles

An unsupervised group detection method for understanding group dynamics in crowds

Pedestrian groups arrive in large numbers in crowd gatherings, especially of a spiritual nature. Various studies have been done on crowd control in public spaces by analysing the behaviour of pedestrian groups. Understanding group dynamics can help better plan pedestrian facilities and large events. Many existing group sensing models primarily determine social bonding between pedestrians using spatiotemporal parameters, such as distance, directional movement, and overlapping time. However, social bonding determined based on these parameters assumes the bonding to be symmetric, spatially and temporally static and is unaffected by neighbourhood. Our study addresses the issue by relaxing such assumptions and developing an unsupervised group detection model based on potential candidates. The proposed model can handle temporal and spatial variations more effectively than those based on simple spatiotemporal parameters. The model developed is assessed both quantitatively and qualitatively. New metrics are introduced for quantitative evaluation, comparing predicted groups and ground truth instead of pedestrian pairs with ground truth. A visualisation method is developed for the qualitative assessment. Group splits and group merges are calculated to assist in understanding crowd movement patterns. Overall, this study helps in further exploring and assessing groups, which can improve understanding of crowd dynamics.

Crowd Density Estimation via Global Crowd Collectiveness Metric

Drone-captured crowd videos have become increasingly prevalent in various applications in recent years, including crowd density estimation via measuring crowd collectiveness. Traditional methods often measure local differences in motion directions among individuals and scarcely handle the challenge brought by the changing illumination of scenarios. They are limited in their generalization. The crowd density estimation needs both macroscopic and microscopic descriptions of collective motion. In this study, we introduce a Global Measuring Crowd Collectiveness (GMCC) metric that incorporates intra-crowd and inter-crowd collectiveness to assess the collective crowd motion. An energy spread process is introduced to explore the related crucial factors. This process measures the intra-crowd collectiveness of individuals within a crowded cluster by incorporating the collectiveness of motion direction and the velocity magnitude derived from the optical flow field. The global metric is adopted to keep the illumination-invariance of optical flow for intra-crowd motion. Then, we measure the motion consistency among various clusters to generate inter-crowd collectiveness, which constitutes the GMCC metric together with intra-collectiveness. Finally, the proposed energy spread process of GMCC is used to merge the inter-crowd collectiveness to estimate the global distribution of dense crowds. Experimental results validate that GMCC significantly improves the performance and efficiency of measuring crowd collectiveness and crowd density estimation on various crowd datasets, demonstrating a wide range of applications for real-time monitoring in public crowd management.

Robust Estimation of Crowd Density Using Vision Transformers

Estimating and predicting crowd density at large events or during disasters is of paramount importance for enhancing emergency management systems. This includes planning effective evacuation routes, optimizing rescue operations, and ensuring efficient deployment of emergency services. Traditionally, surveillance systems that rely on cameras have been employed to monitor crowd movements. However, accurately estimating crowd density using such systems presents several challenges. These challenges stem primarily from the interaction between large crowds and the limitations of two-dimensional cameras in capturing the full scope of three-dimensional spaces. Optical distortions, environmental factors, and variations in camera angles further complicate the task, making accurate estimations difficult to achieve. To address these challenges, this paper introduces a robust method for calculating crowd density that leverages advanced vision transformers. By combining the output of these transformers with a two-stage neural network, the method effectively mitigates the limitations of traditional approaches. One of the key advantages of the proposed system is its robustness, which allows it to perform well across different camera specifications, installation locations, and image aspect ratios. The method applies and evaluates various deep learning techniques, introducing improvements to existing network structures that are better suited for the problem at hand. Extensive experimental verification demonstrates that the proposed method consistently produces accurate crowd density estimates, even in diverse and complex crowd environments. This robust performance underscores its potential for improving emergency management and crowd control in real-world situations.

Fire Safety Assessment of a Typical Sports Hall Building Based on Fire Dynamics and Crowd Movement Models

Fire risk analysis is one of the essential components of building design to ensure the safety of occupants and properties in case of fires. Currently, the Ministry of Public Works and Housing Regulation No. 20/PRT/M/2009 provides guidelines for conducting a fire risk analysis, however, without a clear consideration of fire dynamics in the estimation of the fire risk level. In this work, we investigate the fire safety aspects of a typical sports hall buildings by a fire dynamics deterministic model (Fire Dynamics Simulator (FDS) of the National Institute of Science and Technology, USA) and crowd movement model for occupant evacuation (Pathfinder of Thunderhead Engineering). Systematic investigations were made on the effects of the fire growth category and smoke extraction system on the Available Safe Egress Time (ASET). The results of ASET were then compared to the Required Safe Egress Time (RSET) which is obtained from the evacuation model. Our results suggest that ASET decreases exponentially with fire growth rate, especially from slow to medium growth rate. The fire growth rate significantly affects the acceptable fire risk of ASET longer than RSET. Occupant capacity, fire management systems, and smoke extraction systems play important roles in reducing fire risk. However, as the fire growth rate increases, the effects of smoke extraction in maintaining safe conditions diminish. This study provides recommendations to reduce risks to the occupants in case of fire, contributing to the considerations of the design and management of a typical sports hall building.

Temporally enhanced abnormal behavior detection based on multi-channel coupling

Abnormal behavior detection in surveillance videos based on deep learning has becomea hot topic in current research. However, due to the complexity and variability of crowd movement and external environment, detecting abnormal behavior faces great challenges. Current abnormal behavior recognition models have limitations in feature extraction and pay insufficient attention to dynamic temporal features. To address these problems, a spatiotemporal enhancement anomaly detection method based on multi-channel coupling is proposed. Based on the SlowFast network, a multi-channel coupled temporal enhancement module and a spatial enhancement module are introduced respectively to extract more discriminative static and dynamic feature information. A large number of experiments are carried out on three benchmark datasets (Violent Flow, Hockey Fight, and Real-life Violence Situations). The results show that the prediction accuracy of the proposed abnormal behavior recognition method reaches 95.3%、97.3%and respectively 94%, thus verifying the effectiveness of the method.

The effect of building bottlenecks on crowd dynamics involving individuals with simulated disabilities

With the development of urbanization and the growth of population, there is a growing demand for safety in public building facilities. As one of the essential building components of urban architecture, bottlenecks have a significant impact on the evacuation efficiency of crowds. Furthermore, the heterogeneity of crowds also contributes to the complexity of crowd movement through bottlenecks, while aggravating the magnitude of congestion induced by bottlenecks. The objective of this paper is to explore the movement characteristics of heterogeneous crowds passing through a corridor with a bottleneck by conducting a controlled experiment. There were three variables in this experiment, namely the individual categories (i.e., able-bodied individuals, simulated individuals on crutches and simulated wheelchair users), bottleneck width (i.e., 1.2, 1.6 and 2.0 m) and proportion of simulated disabilities in crowds (i.e., 0 %, 5 % and 10 %). Then offset angle, passing efficiency, fundamental diagram, etc., were analyzed. In trials involving simulated individuals on crutches, a higher detouring degree is observed compared to trials involving simulated wheelchair users or mixed groups of two types of simulated disabilities. There is an increase in flow rate induced by increasing the bottleneck width and decreasing the proportion of simulated disabilities. The passing efficiency at the upstream of the bottleneck in all tests is primarily influenced by the bottleneck width, while by the type and proportion of simulated disabilities at the downstream or inside the bottleneck. The findings are intended to complement the dynamic theory of heterogeneous crowds at building bottlenecks, while providing a reference for congestion control of crowds at bottlenecks.

New insights into the merging behavior of pedestrians: Quantification of lane competition level

Merging behavior is a complex crowd movement pattern caused by certain architectural features and has triggered serious crowd accidents worldwide. It is not uncommon to find such merging structures on the floor plans of some underground infrastructures, which deserve more detailed studies on special crowd movement patterns to enhance the safety level. Researchers have demonstrated that the unstable emotions and competitive behaviors of pedestrians during underground evacuation need to be taken into account due to the enclosed environment. However, there are no relevant studies focused on the competition characteristics among pedestrians during the merging process in a quantitative way. The comprehensiveness and systematization of the geometric features of typical merging structures, as well as the way to create the perception of competition and conflict among pedestrians during real emergencies also need to be improved. In this paper, we carried out a series of controlled experiments in symmetric and asymmetric merging structures with different channel widths and incentive levels to investigate crowd merging behavior with competition characteristics. The density and delay time within the merging area, as well as the time interval and lane changes before and after merging are analyzed. By introducing and modifying Simpson’s Diversity Index, we proposed a new method for quantifying the competition level among pedestrians during the merging process. It is concluded that the performance of the merging area is sensitive to the channel widths and the merging layouts, and the negative effect of unbalanced walking priority is more significant under high incentive level. The lane competition level during the merging process is correlated with the effects of various geometric features, movement layouts, and incentive levels. The method proposed in this paper brings new insights into the interpretation of merging scenarios and more studies on pedestrian dynamics. We hope that this study could serve as supplementary work for other modeling or field experiment methods to jointly provide more insights into the crowd management strategies in some underground infrastructures.

A multi-agent motion simulation method for emergency scenario deduction

Simulating crowd motion in emergency scenarios remains a challenge in computer graphics due to crowd heterogeneity and environmental complexity. However, existing crowd simulation methods homogenize the agent model and simplify target selection and motion navigation of emergency crowds. To address these problems, we propose a multi-agent motion simulation method for emergency scenario deduction. First, we propose a multi-agent model to simulate crowd heterogeneity. This model includes a personality-based heterogeneous agent model and an agent perception model that considers vision, hearing, and familiarity with the environment. Second, we propose a target selection strategy based on the motion patterns of actual pedestrians. This strategy employs mathematical models and our agent perception model to guide agents in selecting appropriate targets. Finally, we propose a global navigation algorithm that combines random sampling with heuristic search methods. Concurrently, we use our multi-agent model to adjust the agent’s local motion planning to deduce the motion states of emergency crowds naturally. Experimental results validate that our method can realistically and reasonably simulate crowd motion in emergency scenarios.

Characteristics of crowd disaster: Database construction and pattern identification

Crowd disasters pose a serious societal threat with frequent occurrence and high fatality. Despite thorough research having been conducted on the basic mechanisms behind crowd disasters, further qualitative study is required to explain patterns of crowd movement under different environmental conditions. This paper describes the methodology behind the creation of a novel crowd disaster database, incorporating unprecedentedly strict inclusion criteria and well-defined characteristics in order to facilitate better qualitative understanding. The data characteristics are retrieved from previous literature reviews and a 2014 Shanghai crowd disaster case study. Online news database Factiva is used as the main data source, combining with supplemental Internet sources. The completed database contains 293 crowd disasters from 1989 to 2021 and is subject to a data quality assessment procedure. Through the cluster analysis algorithm K-Modes, the entire database has been categorized into nine clusters and distinct features within each cluster are discussed. Each cluster is thus considered as a unique type of crowd disaster, enabling event organizers to link combination of characteristics with likelihood of crowd disaster. Consequently, a more explicit crowd management strategy can be developed to target most common causal factors.

Floor Vibrations of Steel-Framed Floor Systems Due to Rhythmic Activities—Case Study

Human induced vibrations on steel-framed floor systems can result in a wide variety of serviceability problems, ranging from discomfort of building occupants to damage of building components. Rhythmic activities, like aerobics, have the potential to cause significant floor vibrations owing to synchronized crowd movement. This article presents the results of a case study in which the susceptibility of a steel-framed floor system to structural vibrations was evaluated after an aerobic dance event caused damage to non-structural building components. The study assessed a floor supported by a system of long span trusses, which at one end were themselves supported by a two-story deep long span truss. This study consisted of collecting onsite vibration measurements and evaluating the floor system using a numerical model. The finite element analysis model was calibrated using results from field measured data. The numerical analysis results showed that estimated floor accelerations exceeded standard limits for different types of rhythmic activities, indicating that the floor system was susceptible to uncomfortable vibrations. Possible methods to improve the structural response of the floor and to address human discomfort are discussed.

Statistical and density-based clustering of geographical flows for crowd movement patterns recognition

With the rapid development of sensors and communication technologies, it has become easy to collect large-scale and long-term crowd movement positioning data, which brings new opportunities for studying crowd movement patterns. This paper proposes a novel Statistical and Density-Based Clustering algorithm (SDBC) to identify implicit significant spatial aggregation patterns in geographical flow data. Unlike existing flow clustering algorithms, this method identifies the hot spots of origin-destination (OD) flows based on local spatial statistics and density-growing clustering. It also evaluates the significance of the identified geographic flow clusters, effectively reducing the identification of spurious clusters generated by chance in data. In our method, the spatial neighborhood of each flow is first obtained based on spatial proximity, temporal similarity, and directional similarity. Then, the number of flows in the spatial neighborhood of each flow is calculated and used as the density measure. Based on this, high-density flows are automatically detected using local spatial aggregation statistics, and a hierarchical density-based clustering strategy is developed to merge adjacent high-density flows to generate candidate flow clusters. Finally, we perform permutation tests to infer the statistical significance of each flow cluster and eliminate the candidate clusters generated by chance. Experiments on synthetic data and real-world taxi trajectory data were conducted to evaluate the effectiveness of the proposed method. Results show that the proposed method can accurately identify the statistically significant flow clusters of different shapes and densities and performs better than the available state-of-the-art flow clustering algorithms.

Evacuation scenario optimization in buildings with human anthropometric characteristics

Emergency evacuations are critical situations that threaten people's lives and safety, necessitating robust planning and prevention strategies. Anthropometric data, which are numerical expressions of human body measurements, play a vital role in architectural design by helping determine safe evacuation routes. This data is crucial for understanding the physical characteristics and mobility of building occupants, allowing for optimized evacuation routes that accommodate the needs of various groups, such as the elderly, children, and individuals with physical disabilities. The Social Forces Model (SFM) is an important tool used to understand the behavior and movement of crowds, modeling these aspects mathematically by considering social interactions and environmental factors. This study develops a system to enhance emergency evacuation scenarios in buildings using anthropometric data and SFM. This system analyzes the evacuation process and identifies optimal routes, contributing significantly to ensuring safety and minimizing damage during disasters. The research shows that using anthropometric data and SFM in architectural design is essential for effective emergency evacuation planning and management. Simulations that incorporate the diverse physical characteristics of people demonstrate notable improvements in evacuation times compared to those using fixed values, with up to 10.20 % improvement in single-exit scenarios and up to 13.45 % in two-exit scenarios. Furthermore, evacuation routes are optimized using SFM, and histograms are employed to identify the most preferred routes during panic situations. This data-driven approach informs the design of building and floor layouts, as well as the materials and decorations used, ensuring that evacuation routes remain clear and accessible. Removing obstacles and facilitating easy passage during emergencies significantly enhances the effectiveness of evacuations. Overall, the integration of anthropometric data and SFM in building design is shown to create faster and safer evacuation processes, ultimately improving emergency response and safety.

Together Apart: the Influence of Increased Crowd Heterogeneity on Crowd Dynamics at Bottlenecks

Individual differences in mobility (e.g., due to wheelchair use) are often ignored in the prediction of crowd movement. Consequently, engineering tools cannot fully describe the impact of vulnerable populations on egress performance. To contribute to closing this gap, we performed laboratory experiments with 25 pedestrians with varying mobility profiles. The control condition comprised only participants without any additional equipment; in the luggage condition and the wheelchair condition, two participants at the center of the group either carried suitcases or used a wheelchair. We found that individuals using wheelchairs and to a lesser degree those carrying luggage needed longer to pass through the bottleneck, which also affected those walking behind them. This led to slower times to fully clear the bottleneck in the wheelchair and luggage condition compared to the control group. The results challenge the status quo in existing approaches to calculating egress performance and other key performance metrics in crowd dynamics.

Vision-based survey method for extraordinary loads on buildings

The statistical modeling of extraordinary loads on buildings has been stagnant for decades due to the laborious and error-prone nature of existing survey methods, such as questionnaires and verbal inquiries. This study proposes a new vision-based survey method for collecting extraordinary load data by automatically analyzing surveillance videos. For this purpose, a crowd head tracking framework is developed that integrates crowd head detection and reidentification models based on convolutional neural networks to obtain head trajectories of the crowd in the survey area. The crowd head trajectories are then analyzed to extract crowd quantity and velocities, which are the essential factors for extraordinary loads. For survey areas with frequent crowd movements during temporary events, the equivalent dynamic load factor can be further estimated using crowd velocity to consider dynamic effects. A crowd quantity investigation experiment and a crowd walking experiment are conducted to validate the proposed survey method. The experimental results prove that the proposed survey method is effective and accurate in collecting load data and reasonable in considering dynamic effects during extraordinary events. The proposed survey method is easy to deploy and has the potential to collect substantial and reliable extraordinary load data for determining design load on buildings.

Crowd Movement Type Estimation in Video by Integral Optical Flow and Convolution Neural Network

Crowd Movement Type Estimation in Video by Integral Optical Flow and Convolution Neural Network

A Novel Movable Mannequin Platform for Evaluating and Optimising mmWave Radar Sensor for Indoor Crowd Evacuation Monitoring Applications

Developing mmWave radar sensors for indoor crowd motion sensing and tracking faces a critical challenge: the scarcity of large-scale, high-quality training data. Traditional human experiments encounter logistical complexities, ethical considerations, and safety issues. Replicating precise human movements across trials introduces noise and inconsistency into the data. To address this, this study proposes a novel solution: a movable platform equipped with a life-size mannequin to generate realistic and diverse data points for mmWave radar training and testing. Unlike human subjects, the platform allows precise control over movements, optimising sensor placement relative to the target object. Preliminary optimisation results reveal that sensor height impacts tracking performance, with an optimal sensor placement above the test subject yields the best results. The results also reveal that the 3D data format outperforms 2D data in accuracy despite having fewer frames. Additionally, analysing height distribution using 3D data highlights the importance of the sensor angle—15° downwards from the horizontal plane.

Learning Crowd Motion Dynamics with Crowds

Reinforcement Learning (RL) has become a popular framework for learning desired behaviors for computational agents in graphics and games. In a multi-agent crowd, one major goal is for agents to avoid collisions while navigating in a dynamic environment. Another goal is to simulate natural-looking crowds, which is difficult to define due to the ambiguity as to what is a natural crowd motion. We introduce a novel methodology for simulating crowds, which learns most-preferred crowd simulation behaviors from crowd-sourced votes via Bayesian optimization. Our method uses deep reinforcement learning for simulating crowds, where crowdsourcing is used to select policy hyper-parameters. Training agents with such parameters results in a crowd simulation that is preferred to users. We demonstrate our method's robustness in multiple scenarios and metrics, where we show it is superior compared to alternate policies and prior work.

Dynamics characteristic of pedestrians’ particular overtaking behavior based on an improved social force model

The overtaking behavior of individuals under anxious panic is a major causative factor of crowded trampling accidents. In order to simulate the overtaking behavior of individuals experiencing anxiety and panic in a single-exit room within a normal pedestrian flow, an improved social force model is proposed to simulate crowd movement at local medium and high density through introducing the bypass overtaking mechanism, the side overtaking mechanism, and the side avoidance mechanism generated by association. The numerical simulation verifies the model’s validity. The effect of the size and number of overtaking individuals on the pedestrian flow are explored. The results show that the new model can simultaneously reproduce a variety of overtaking behaviors and their associated behaviors in line with reality. The larger size of the overtaking individuals are, the easier it is to produce overtaking behaviors and the more difficult it is to overtake. With the increase of the number of overtaking individuals, the evacuation time becomes longer, but there is no linear relationship between them. When the initial pedestrian density is 0.37ped/m2, with time evolution, the appearance frequency of local density greater than 0.8ped/m2 suddenly increases, and the comfort and safety of pedestrian movement decreases. The results of this paper can enrich the social force simulation model of pedestrian flow and provide a theoretical guidance for the evacuation of pedestrian flow with a few overtaking individuals in the room.

Experimental study on individual and crowd movement features around obstacles with different shape and size

Well-structured pedestrian experiments in a long corridor have been carried out to explore the influence of obstacle shape and size on individual and crowd level pedestrian movement characteristics. Results indicate that for individual pedestrians, the number of right-turning pedestrians and the target drift angle show clear changes with the increase of the obstacle size, while the speed only changes significantly when the obstacle size is greater than 1.5 m. For the crowd movement scenarios, a small obstacle can speed up the pedestrian flow, then, with the increase of the obstacle size, the movement time increases. The increase rate has a relation with the obstacle shape. The obstacle shape influence becomes more obvious when the individual and crowd movement scenarios are compared. The results of this paper are expected to provide practical basis for modeling pedestrian under the influence of obstacle.

Efficient Crowd Anomaly Detection Using Sparse Feature Tracking and Neural Network

Crowd anomaly detection is crucial in enhancing surveillance and crowd management. This paper proposes an efficient approach that combines spatial and temporal visual descriptors, sparse feature tracking, and neural networks for efficient crowd anomaly detection. The proposed approach utilises diverse local feature extraction methods, including SIFT, FAST, and AKAZE, with a sparse feature tracking technique to ensure accurate and consistent tracking. Delaunay triangulation is employed to represent the spatial distribution of features in an efficient way. Visual descriptors are categorised into individual behaviour descriptors and interactive descriptors to capture the temporal and spatial characteristics of crowd dynamics and behaviour, respectively. Neural networks are then utilised to classify these descriptors and pinpoint anomalies, making use of their strong learning capabilities. A significant component of our study is the assessment of how dimensionality reduction methods, particularly autoencoders and PCA, affect the feature set’s performance. This assessment aims to balance computational efficiency and detection accuracy. Tests conducted on benchmark crowd datasets highlight the effectiveness of our method in identifying anomalies. Our approach offers a nuanced understanding of crowd movement and patterns by emphasising both individual and collective characteristics. The visual and local descriptors facilitate high-level analysis by closely relating to semantic information and crowd behaviour. The analysis observed shows that this approach offers an efficient framework for crowd anomaly detection, contributing to improved crowd management and public safety. The proposed model achieves accuracy of 99.5 %, 96.1%, 99.0% and 88.5% in the UMN scenes 1, 2, and 3 and violence in crowds datasets, respectively.