Crowd density estimation using deep learning for Hajj pilgrimage video analytics

Background: This paper focuses on advances in crowd control study with an emphasis on high-density crowds, particularly Hajj crowds. Video analysis and visual surveillance have been of increasing importance in order to enhance the safety and security of pilgrimages in Makkah, Saudi Arabia. Hajj is considered to be a particularly distinctive event, with hundreds of thousands of people gathering in a small space, which does not allow a precise analysis of video footage using advanced video and computer vision algorithms. This research proposes an algorithm based on a Convolutional Neural Networks model specifically for Hajj applications. Additionally, the work introduces a system for counting and then estimating the crowd density. Methods: The model adopts an architecture which detects each person in the crowd, spots head location with a bounding box and does the counting in our own novel dataset (HAJJ-Crowd). Results: Our algorithm outperforms the state-of-the-art method, and attains a remarkable Mean Absolute Error result of 200 (average of 82.0 improvement) and Mean Square Error of 240 (average of 135.54 improvement). Conclusions: In our new HAJJ-Crowd dataset for evaluation and testing, we have a density map and prediction results of some standard methods.

Simulation study of crowd dynamics in pedestrian evacuation concerning panic contagion: A cellular automaton approach

The study of the panic evacuation process is of great significance to emergency management. Panic not only causes negative emotions such as irritability and anxiety, but also affects the pedestrians decision-making process, thereby inducing the abnormal crowd behavior. Prompted by the epidemiological SIR model, an extended floor field cellular automaton model was proposed to investigate the pedestrian dynamics under the threat of hazard resulting from the panic contagion. In the model, the conception of panic transmission status (PTS) was put forward to describe pedestrians' behavior who could transmit panic emotions to others. The model also indicated the pedestrian movement was governed by the static and hazard threat floor field. Then rules that panic could influence decision-making process were set up based on the floor field theory. The simulation results show that the stronger the pedestrian panic, the more sensitive pedestrians are to hazards, and the less able to rationally find safe exits. However, when the crowd density is high, the panic contagion has a less impact on the evacuation process of pedestrians. It is also found that when the hazard position is closer to the exit, the panic will propagate for a longer time and have a greater impact on the evacuation. The results also suggest that as the extent of pedestrian's familiarity with the environment increases, pedestrians spend less time to escape from the room and are less sensitive to the hazard. In addition, it is essential to point out that, compared with the impact of panic contagion, the pedestrian's familiarity with environment has a more significant influence on the evacuation.

DGG: A Novel Framework for Crowd Gathering Detection

Crowd gathering detection plays an important role in security supervision of public areas. Existing image-processing-based methods are not robust for complex scenes, and deep-learning-based methods for gathering detection mainly focus on the design of the network, which ignores the inner feature of the crowd gathering action. To alleviate such problems, this work proposes a novel framework Detection of Group Gathering (DGG) based on the crowd counting method using deep learning approaches and statistics to detect crowd gathering. The DGG mainly contains three parts, i.e., Detecting Candidate Frame of Gathering (DCFG), Gathering Area Detection (GAD), and Gathering Judgement (GJ). The DCFG is proposed to find the frame index in a video that has the maximum people number based on the crowd counting method. This frame means that the crowd has gathered and the specific gathering area will be detected next. The GAD detects the local area that has the maximum crowd density in a frame with a slide search box. The local area contains the inner feature of the gathering action and represents that the crowd gathering in this local area, which is denoted by grid coordinates in a video frame. Based on the detected results of the DCFG and the GAD, the GJ is proposed to analyze the statistical relationship between the local area and the global area to find the stable pattern for the crowd gathering action. Experiments based on benchmarks show that the proposed DGG has a robust representation of the gathering feature and a high detection accuracy. There is the potential that the DGG can be used in social security and smart city domains.

Unmanned Aerial Vehicles for Crowd Monitoring and Analysis

Crowd monitoring and analysis has become increasingly used for unmanned aerial vehicle applications. From preventing stampede in high concentration crowds to estimating crowd density and to surveilling crowd movements, crowd monitoring and analysis have long been employed in the past by authorities and regulatory bodies to tackle challenges posed by large crowds. Conventional methods of crowd analysis using static cameras are limited due to their low coverage area and non-flexible perspectives and features. Unmanned aerial vehicles have tremendously increased the quality of images obtained for crowd analysis reasons, relieving the relevant authorities of the venues’ inadequacies and of concerns of inaccessible locations and situation. This paper reviews existing literature sources regarding the use of aerial vehicles for crowd monitoring and analysis purposes. Vehicle specifications, onboard sensors, power management, and an analysis algorithm are critically reviewed and discussed. In addition, ethical and privacy issues surrounding the use of this technology are presented.

Crowd density estimation using deep learning for Hajj pilgrimage video analytics

Background: This paper focuses on advances in crowd control study with an emphasis on high-density crowds, particularly Hajj crowds. Video analysis and visual surveillance have been of increasing importance in order to enhance the safety and security of pilgrimages in Makkah, Saudi Arabia. Hajj is considered to be a particularly distinctive event, with hundreds of thousands of people gathering in a small space, which does not allow a precise analysis of video footage using advanced video and computer vision algorithms. This paper aims to propose an algorithm based on a Convolutional Neural Networks model specifically for Hajj applications. Additionally, the work introduces a system for counting and then estimating the crowd density. Methods: The model adopts an architecture which detects each person in the crowd, spots head location with a bounding box and does the counting in our own novel dataset (HAJJ-Crowd). Results: Our algorithm outperforms the state-of-the-art method, and attains a remarkable Mean Absolute Error result of 200 (average of 82.0 improvement) and Mean Square Error of 240 (average of 135.54 improvement). Conclusions: In our new HAJJ-Crowd dataset for evaluation and testing, we have a density map and prediction results of some standard methods.

A Quantitative Approach of Subway Station Passengers’ Heterogeneity of Decision Preference Considering Personality Traits During Emergency Evacuation

Subway station emergencies are gradually increasing in China. The aim of this research is to study the effects of “Dist”, “Pedestrian flow” and “Crowd density” on the heterogeneity of passengers’ decision-making preference and explore the relationship between heterogeneity and personality. Firstly, a questionnaire of 20 emergency evacuation scenarios, that includes the Eysenck Personality Questionnaire, is designed. Secondly, the heterogeneity of passengers’ decision preference is quantified by the random parameter logit model. Finally, personality traits and influencing factors are used as abscissa and ordinate respectively, to study the relationship between personality traits and preference heterogeneity. The results show that the coefficients of “Dist”, “Pedestrian flow” and “Crowd density” are –0.101, 0.236 and –0.442 respectively, which are statistically significant. The proportion of extroverted passengers of the exit is 9% higher than that of introverted passengers when “Pedestrian flow” of the exit is greater than the average value, while the proportion of introverted passengers is 7% higher than that of extroverted passengers when “Crowd density” is smaller than the average value. The conclusion is that the three influence factors are random variables, and “Dist” shows the lowest level of heterogeneity. Extroverted passengers are more likely to follow a large crowd for evacuation, but introverted passengers are more likely to avoid crowded exits.

Multiscale Aggregate Networks with Dense Connections for Crowd Counting

The most advanced method for crowd counting uses a fully convolutional network that extracts image features and then generates a crowd density map. However, this process often encounters multiscale and contextual loss problems. To address these problems, we propose a multiscale aggregation network (MANet) that includes a feature extraction encoder (FEE) and a density map decoder (DMD). The FEE uses a cascaded scale pyramid network to extract multiscale features and obtains contextual features through dense connections. The DMD uses deconvolution and fusion operations to generate features containing detailed information. These features can be further converted into high-quality density maps to accurately calculate the number of people in a crowd. An empirical comparison using four mainstream datasets (ShanghaiTech, WorldExpo’10, UCF_CC_50, and SmartCity) shows that the proposed method is more effective in terms of the mean absolute error and mean squared error. The source code is available at https://github.com/lpfworld/MANet.

Spatial analysis for crowds in multi-directional flows based on large-scale experiments

In this paper, spatial analysis for the nearest neighbors is performed in the unidirectional, bidirectional and crossing flows. Based on the intended direction given in the experiment, different types of neighbors such as U-ped (neighbor with the same intended direction), B-ped (neighbor with the opposite intended direction) and C-ped (neighbor with the intersecting intended direction) are defined. The preferable positions of these neighbors during movement are investigated under various conditions. The spatial relation is quantified by calculating the distance and angle between the reference pedestrian and neighbors. The results indicate that the distribution of neighbors is closely related to the neighbor’s order, crowd density, neighbor type and flow type. Furthermore, the reasons that result in these distributions for different neighbors are explored. Finally neighbor distributions for different flows are compared and the implications of this research are discussed. The spatial analysis sheds new light on the study of pedestrian dynamics in a different perspective, which can help to develop and validate crowd models in the future.

One-Way Coupling of Fire and Egress Modeling for Realistic Evaluation of Evacuation Process

In the discipline of fire engineering, computational simulation tools are used to evaluate the available safe egress time (ASET) and required safe egress time (RSET) of a building fire. ASET and RSET are often analyzed separately, using computational fluid dynamics (CFD) and crowd dynamics, respectively. Although there are advantages to coupling the ASET and RSET analysis to quantify tenability conditions and reevaluate evacuation time within a building, the coupling process is computationally complex, requiring multiple steps. The coupling setup can be time-consuming, particularly when the results are limited to the modeled scenario. In addition, the procedure is not uniform throughout the industry. This paper presents the successful one-way coupling of CFD and crowd dynamics modeling through a new simplified methodology that captures the impact of fractional effective dose (FED) and reduced visibility from smoke on the individual evacuee’s movement and the human interaction. The simulation tools used were Fire Dynamics Simulator (FDS) and Oasys MassMotion for crowd dynamics. The coupling was carried out with the help of the software development kit of Oasys MassMotion in two different example geometries: an open-plan room and a floor with six rooms and a corridor. The results presented in this paper show that, when comparing an uncoupled and a coupled simulation, the effects of the smoke lead to different crowd density profiles, particularly closer to the exit, which elongates the overall evacuation time. This coupling method can be applied to any geometry because of its flexible and modular framework.