Crowd flow forecasting via agent-based simulations with sequential latent parameter estimation from aggregate observation

The advent of the COVID-19 pandemic has adversely affected the entire world and has put forth high demand for techniques that remotely manage crowd-related tasks. Video surveillance and crowd management using video analysis techniques have significantly impacted today's research, and numerous applications have been developed in this domain. This research proposed an anomaly detection technique applied to Umrah videos in Kaaba during the COVID-19 pandemic through sparse crowd analysis. Managing the Kaaba rituals is crucial since the crowd gathers from around the world and requires proper analysis during these days of the pandemic. The Umrah videos are analyzed, and a system is devised that can track and monitor the crowd flow in Kaaba. The crowd in these videos is sparse due to the pandemic, and we have developed a technique to track the maximum crowd flow and detect any object (person) moving in the direction unlikely of the major flow. We have detected abnormal movement by creating the histograms for the vertical and horizontal flows and applying thresholds to identify the non-majority flow. Our algorithm aims to analyze the crowd through video surveillance and timely detect any abnormal activity to maintain a smooth crowd flow in Kaaba during the pandemic.

This paper describes a solution for monitoring and detection of crowds and analysis of density structures and movement characteristics, to enhance safety of citizens and security of critical infrastructures. The system leverages the Internet of Things concept and heterogenous, energy efficient, networked sensors, with support for wireless communication. Privacy protection, instant deployability and auto configuration are hereby underlying core objectives. The solution, which will be described, comprises two novel distributed crowd analysis algorithms, allowing on the one hand the localisation of critical areas within large crowds and on the other hand the recognition of counter streams, which can cause severe impacts on the crowd flow and movement velocity and which can transform crowding scenarios into threatening situations.

Abstract: WiFiTally is a cutting-edge technology designed for monitoring and counting crowds in public spaces using Wi-Fi signals. Imagine a system that can track and analyze the movement of people in areas like shopping malls, airports, or event venues without invading privacy. WiFiTally achieves this by tapping into the Wi-Fi signals emitted by smartphones and other connected devices carried by individuals. In simple terms, WiFiTally observes the patterns of Wi-Fi signals within a designated area, providing insights into crowd density and movement. Instead of relying on traditional methods like cameras, this technology prioritizes privacy by using existing Wi-Fi signals that people willingly emit from their devices. The process involves collecting anonymous data from Wi-Fi signals and transforming it into valuable information about crowd size and flow. It doesn't identify individuals; instead, it focuses on overall trends, helping businesses and event organizers make informed decisions about crowd management. WiFiTally offers a non-intrusive and privacy-friendly solution for understanding and optimizing crowd dynamics. This abstract aims to introduce the concept of WiFiTally crowd monitoring and counting, emphasizing its user-friendly approach and respect for personal privacy in public spaces.

Simulating crowds is a challenging but important problem. There are various methodologies in the literature ranging from macroscopic numerical flow simulations to detailed, microscopic agent simulations. One key issue for all crowd simulations is scalability. Some methods address this issue through abstraction, describing global properties of homogeneous crowds. However, ideally a modeler should be able to simulate large heterogeneous crowds at fine levels of detail. We are attempting to achieve scalability through the application of distributed simulation techniques to agent-based crowd simulation. Generally, for an agent-based crowd simulation that aims to simulate a large number of agents under different scenarios, a proper partitioning algorithm is an essential factor to the performance of the system. However, selecting/developing an appropriate partitioning algorithm for a specific agent-based crowd simulation still remains an issue. One way is to implement distributed agent-based crowd simulations using multiple partitioning algorithms and to compare their performance. This requires multiple distributed simulation experiments, one for each partitioning algorithm. Considering the resources and the effort needed to perform multiple distributed simulation experiments, it is therefore preferred to predict the performance of partitioning algorithms instead. The research reported in this thesis focuses on investigating an effective and efficient approach to evaluate proposed partitioning algorithms for agent-based crowd simulations. A generic framework and an evaluation process are proposed to address the issue of selecting an appropriate partitioning algorithm by simulating the effect of a partitioning algorithm i Firstly, I would like to render my heartfelt appreciation to my mentor,

Smart surveillance systems developed in recent years have made enormous contributions to providing safety and management of crowds. The aim of this study is to observe and try to understand how crowd movements presented in a video sequence show behaviour. For this end, the motion data at pixel level among the consecutive frames is obtained using optical flow initially. Then, this motion data is associated using the particle advection method and stable as well as moving areas in the image are obtained. After, the moving areas clustered using Mean-Shift method are described and classified as parabola, in addition to the studies in the literature. At the end of the study, the method developed was tested over UCF as well as Pets2009 datasets and the results are presented.

Technical advancements allow for the development of crowd monitoring and management support systems to ensure the safety of pedestrians during large-scale events. This way, pedestrian behaviour can be monitored during events, and potential dangerous situations can be identified in a timely manner. A real-time prototype Crowd Monitoring Dashboard has been developed for a large nautical event in Amsterdam in 2015. Three main functions of such a system have been identified: the real-time data collection, the traffic engineering functions and the visualization. This paper focuses on three state estimation algorithms in the second component - the traffic engineering functions. These algorithms are presented and cross compared in terms of performance, data usage and practical applications, based on the empirical data collected from this large-scale event. The outcome can shed useful light on the feasibility of the data sources and estimation methods for future real-time applications on crowd monitoring and management for large-scale events.

On guaranteed parameter estimation of a multiparameter linear regression process

This thesis investigates the use of modelling and simulation techniques in urban areas of smart cities, also exploring how big data can be used to feed these models. These modelling techniques have been applied to two different fields that have been gaining prominence during the last years but where research is still limited: urban logistics and urban resilience. Through this thesis, the author has expanded the research knowledge in these fields by exploring different methods such as meta-heuristics, transport modelling, and agent-based simulation in order to define new methodologies to be applied to urban areas. Regarding logistics, the author has shown through the use of meta-heuristics that when traffic congestion is considered as a dynamic attribute to optimize delivery routes in urban areas, time can be reduced by 11%, which is crucial for logistics companies in a market that is fiercer every day. This is true not only for urban areas, but this research has also demonstrated that optimizing routes with dynamic congestion attributes is also beneficial at a strategic level for routes between cities. To consider congestion costs in real time, a new approach has been developed in which data from Google is downloaded to feed these meta-heuristic models, although other sources of big data could be also used. In this thesis, a methodology is also presented that has been used to model logistics routes in urban areas considering real-time data and with the flexibility to add different network attributes (gradient, traffic bans, CO2, etc.) to simulate different scenarios. This can be useful for logistics companies to optimize their deliveries (choosing between van or tricycles, selecting the time of the day to deliver, etc.) but also for public authorities to get guidance on different transport and urban policies (pedestrianization of some streets, traffic bans, etc.).As for city resilience, the thesis focuses on evacuation planning. A new methodology has been created in which agent-based simulation is used through interconnected sub-models to model a large-scenario evacuation scenario (flooding event as a consequence of a dam collapse). This research defines the data needed to create these models that can be of great help to improve city resilience, and also analyzes how traffic congestion can affect the evacuation procedures. Through the different research articles that compose this thesis, the author brings light to these fields by developing new methodologies and using real case-studies that can help urban planners, companies, and policy makers to create more efficient, sustainable, and resilient smart cities.

This work proposes a learning-based approach for complex environments, focusing on crowd management in smart cities and the monitoring of public areas through aerial images. The proposed Siamese Like architecture features two parallel input branches, combining a ResNet18 encoder and a U-Netinspired decoder to integrate density maps and crowd count estimations. The main innovations lie in: (i) the use of a dual-input Siamese Like structure to enhance feature learning; (ii) the inclusion of a Refiner Fully Connected (Refiner FC) module that fine-tunes the predicted count through residual learning; (iii) the combination of synthetic (GTA) and real-world (ShanghaiTech Part A) datasets to improve generalization; and iv) implementation of the architecture on edge computing board. The proposed model achieved a Mean Absolute Error (MAE) of 76.16 on the ShanghaiTech Part A dataset, showing strong performance in dense and challenging crowd scenarios.

Automatic visual monitoring systems gain more significance as the enhancement of high speed processors that are able to execute the prolonged algorithm efficiently. Video surveillance monitoring systems are typically applied to study human behavior in a crowded situation. It is used to observe posture, movement, trajectories and the interaction between people. For this purpose CCD cameras are installed at many crowded places. Most researchers’ work however is focused on the people classification and the recognition of activities. Human identification is important because of a need of surveillance in the crowded scenes. Group motion tracking is applied to monitor the Groups of People (GOP) as single entities in the global flow rather than monitoring each person individually. It has been studied on the basis of dynamic movement and the surrounding of crowd density which will affect the speed of an individual/group. Large crowd flow perhaps causes deadly accidents if large numbers of GOP smash into each other in narrow passageways, streets or bridges. The system must identify multiple GOP within their global flow according to their Spatio-temporal characteristics. Figure 1 illustrates the density of large crowd flow. The Religious and Sport activities are the application examples. According to Crowd Dynamics Ltd, (Crowd Dynamics Ltd) in the year of 2006, 363 pilgrims lost their lives while performing the Hajj. There are a number of readily available wide area surveillance systems but most of the systems are not suitable for high density crowd monitoring. For safety surveillance of large crowd, there should be a system which can detect and track unstable GOP by careful monitoring of crowd and display results in realtime scenario. And, with the results being displayed in real-time, it will certainly assist the system’s operator to make better decision.

We report work on speech enhancement that combines sequential noise estimation and perceptual filtering. The sequential estimation employs an extension of the sequential EM-type algorithm. In the algorithm, statistics of clean speech are modeled by hidden Markov models (HMM) and noise is assumed to be Gaussian distributed with a time-varying mean vector (the noise parameter) to be estimated. The estimation process uses a non-linear function that relates speech statistics, noise, and noisy observation. With the estimated noise parameter, the subtraction-type algorithm for speech enhancement may be extended to non-stationary environments. In particular, a perceptual filter with frequency masking is constructed with a tradeoff between noise reduction and speech distortion considering the sensitivity of speech recognition systems to speech distortion. Our experiments in speech enhancement and speech recognition in non-stationary noise confirmed that this approach seems promising in improving performances compared to alternative speech enhancement algorithms.

The escalating process of urbanization has raised concerns about incidents arising from overcrowding, necessitating a deep understanding of large human crowd behavior and the development of effective crowd management strategies. This study employs computational methods to analyze real-world crowd behaviors, emphasizing self-organizing patterns. Notably, the intersection of two streams of individuals triggers the spontaneous emergence of striped patterns, validated through both simulations and live human experiments. Addressing a gap in computational methods for studying these patterns, previous research utilized the pattern-matching technique, employing the Nelder-Mead Simplex algorithm for fitting a two-dimensional sinusoidal function to pedestrian coordinates. This paper advances the pattern-matching procedure by introducing Simulated Annealing as the optimization algorithm and employing a two-dimensional square wave for data fitting. The amalgamation of Simulated Annealing and the square wave significantly enhances pattern fitting quality, validated through statistical hypothesis tests. The study concludes by outlining potential applications of this method across diverse scenarios.

Comparative analysis of the performance of humanitarian logistic structures using agent-based simulation

Pedestrian crowd flows in shared spaces: Investigating the impact of geometry based on micro and macro scale measures

Segmenting and analyzing crowd flow from surveillance videos is effective for monitoring abnormal motion or reducing incidents in a crowd scene. In this paper, we use translation flow to approximate local crowd motion and propose a novel region growing scheme to segment crowd flow based on optical flow field. We improve the model of translation domain segmentation and adapt it to a general vector field. To implement flow segmentation, the domain’s contour determined by a set of boundary points is adaptively updated by shape optimization in the improved model. The experiments based on a set of crowd videos show that the proposed approach has the capability to segment crowd flow for further analysis.KeywordsCrowd Flow SegmentationTranslation flowShape OptimizationAdaptive ContourRegion Growing