Modelling social distancing in crowds

Type of work: data analysis, modelling, calibration

Potential partners: INCONTROL

Potential supervisors: Dorine Duives and Martijn Sparnaaij

Pedestrian Dynamics is currently used to evaluate crowd management measures. Given the outbreak of Corona, social distancing has changed the crowd's behaviour. INCONTROL has added parameters to make it possible to model social distancing with PD. However, these parameters have not yet been calibrated. TU Delft has data pertaining the current movement behaviour of people in the 1.5m society. The objective of this research is to calibrate the microscopic pedestrian simulation model pedestrian dynamics with respect to the social distancing rules. Here, the multi-objective framework developed by Martijn Sparnaaij will serve as the basis for the calibration process.The main goal of the calibration is to increase the accuracy of the predictions by determining the set of values for the model that allows for the best replication of reality. The work entails a combination of data analysis, modelling and calibration. The student is expected to perform a literature study, perform the calibration and write a thesis.

Understand pedestrian wayfinding behaviour in complex buildings

Type of work: experimental work, VR, data analysis

Potential partners: -

Potential supervisors: Dorine Duives and Yan Feng

Wayfinding behaviour investigates the processes that take place when pedestrian orient themselves and navigate through space. Almost all studies into pedestrian route and exit behaviour have limited themselves to investigate pedestrian movement in the horizontal levels. As such, literature does not capture the increasing complexity and difficulty when pedestrians move in a multi-level building with both horizontal and vertical movements in a long distance. The object of this project is to conduct field experiments in the Civil engineering and geosciences Faculty to investigate pedestrian wayfinding behaviour both in horizontal and vertical level. Meanwhile, this graduation project will build on previous work of using VR to investigate pedestrian wayfinding behaviour. In that study, various aspects of pedestrian wayfinding behaviour data was collected. 

Development generic waiting location model train platforms

Type of work: Field experiment, data analysis, modelling 

Potential partners: ProRail and/or NS 

Potential supervisors: Winnie Daamen, Dorine Duives or Serge Hoogendoorn

The flows of passengers through a transfer hub is partly determined by their waiting location on the platform. Based on previous research by Flurin Hanseler this location is influenced by many factors, among other things the facilities of the platform, the locations of the entrance and exits, the stopping location of the train and the distribution of other passengers over the platform. The objective of this research is to develop a machine learning or discrete choice model that can simulate the distribution of waiting passengers on a platform. The work entails a combination of data collection featuring waiting behavior, the analysis of the data and the modelling of the behavior. Furthermore, the student is expected to perform a literature study regarding waiting behavior and write a thesis. 

Development of a crowd movement forecasting model using machine learning

Type of work: Big data analysis, modelling 

Potential partners: AMS, Municipality Amsterdam, NS or ID&T 

Potential supervisors: Dorine Duives, Martijn Sparnaaij

Crowds occur frequently in the Netherlands. Crowd monitoring systems are used to analyse these crowd movements in real-time. In order to allow crowd managers to deploy their crowd management measures before accidents occur, a decent prediction of the unfolding of crowd movement dynamics is necessary. The objective of this assignment is to develop a crowd movement forecasting model for either urban traffic, transfer hubs or large-scale events. During this project you will work with the data from the latest crowd monitoring systems and machine learning techniques. 

Modelling cyclist yielding behavior when merging at T-junction

Type of work: cyclists, behavioural analysis, modelling

Potential partners: -

Potential supervisors: Winnie Daamen, Alexandra Gavriilidou

Cyclists are subject to traffic rules just like motorized vehicles but due to their agility, size and speed, as well as the lack of law enforcement, they do not always comply with them. Policy makers and infrastructure designers are interested in and could benefit from understanding the behaviour of cyclists towards rules, so that they can develop and apply appropriate design guidelines. One situation of interest is the yielding behaviour of cyclists merging at a T-junction and the corresponding guideline pertains to the use of lane markings to make the junction safer and more efficient. To assess this, video data has been collected as part of a large-scale cycling experiment. The objective of this project is to model the yielding behaviour of cyclists using as a starting point an existing model.

Impact assessment crowd management decision-support systems

Type of work: stakeholder analysis, data analysis 

Potential partners: city of Amsterdam / city of Barcelona / city of Milan 

Potential supervisors: Dorine Duives

Summer of 2020 and 2021 four living labs featuring crowd management decision-support systems will be operational in three big European cities. Currently, the impact of these systems is unclear. As a result, it is difficult to show the benefits of this new technological advancement. It will be your task to speak with all parties involved and assess their needs, the capabilities of the systems that are in place, and see where this product has the most added value. Your assignment involves travelling to all three cities and speaking to local operators of these systems. 

Testing capabilities of RFID-tag tracking devices for tourist and event management

Type of work: data analysis, pedestrian state estimation 

Partners outside T&P: city of Amsterdam or University of Catalonia (UPC)

Potential supervisors: Dorine Duives, Winnie Daamen or Serge Hoogendoorn

Summers of 2020 and 2021 four living labs featuring crowd management decision-support systems will be operational in three big European cities. Within two of these labs a new crowd sensing technology will be tested. The aim of this project is to determine the capabilities of this new technology, which features a combination of floating RFID-tags and static sensing stations. In the end, the municipalities of Amsterdam and Barcelona want to use this technology to study the movement behaviour of visitors of their city that are difficult to capture using other sensing techniques. In order to do so, an experiment needs to be set up, operationalized in one of the two labs, and the data will need to be analyzed.  

Rotation behaviour of pedestrians in high density bidirectional and crossing streams

Type of work: Data processing, Data analysis 

Partners outside T&P: - 

Potential supervisors: Martijn Sparnaaij and Dorine Duives

Rotating your body is a common method we all use to manoeuvre and squeeze ourselves through a dense crowd. However, even though this behaviour is common, many questions such as; Howe often do we use body rotation as a collision avoidance strategy?; or; Under what circumstances and densities do we rotate our bodies? are unanswered. The objective of this assignment is to gain more insights into the rotation behaviour of pedestrians using the data from the CrowdLimits experiments. During the project you will develop an algorithm, based on computer vision techniques and machine learning, that automatically extracts the body rotation from the video data and use this to obtain multiple data sets. You can then use these data sets to test multiple hypotheses about rotation behaviour of pedestrians in high density flows. The assignment also entails doing a literature review on rotation behaviour and the writing of a thesis. 

Validating cyclist yielding model at X-intersections

Type of work: data analysis, modelling 

Partners outside T&P: - 

Potential supervisors: Winnie Daamen, Alexandra Gavriilidou

Recently, a model that describes cyclist yielding behaviour has been developed within the AMlab. The model follows the framework introduced in Gavriilidou et al. (2019). The question is how accurate and generic this model is. The objective of this project is to validate this model using data from another intersection. These data come from an already performed cycling experiment in Ahoy, and the trajectories are readily available. 

Forecasting crowdedness in urban environments

Type of work: data analysis, forecasting 

Potential partners: Municipality of Amsterdam 

Potential supervisors: Dorine Duives or Winnie Daamen

The municipality of Amsterdam is overrun with tourists. In order to manage the situatie they deploy hosts, security personnel and cleaning services. In order to estimate how much personnel they need to deploy and to identify the most effective crowd management strategies, insight in the expected number of people is needed. In the last year, the municipality has developed a tool called 'DrukteRadar' using historic information on crowdedness.  Your task will be to figure out to what extent the predictions of 'DrukteRadar' are realistic and whether the predictions by this tool can be improved by means of another forecasting technique or additional factors that might explain the crowdedness. 

TU Delft Library in 1.5-m society

Type of work: data analysis, modelling, design 

Potential partners: - 

Potential supervisors: Yufei Yuan, Winnie Daamen 

The coronavirus has caused one of the biggest worldwide crises since the modern age. Many countries have taken measures to reduce the spreading of the virus. In the Netherlands the government has chosen for a so-called ‘intelligent lockdown’. This means that in general people should stay inside, but they are still free to go outside, on the condition of keeping at least 1.5 meter distance from each other.  

TU Delft Library is one of the main providers of study space (seats) for students and staff members. When government measures allow, this place will be reopening. It is expected that visitors should still keep 1.5 meter distance from each other. Dedicated measures should be designed to facilitate the distance keeping, and monitoring may support to evaluate the efficiency of these measures.   

We have come up with four research topics related to the crowd modelling, monitoring and management for TU Delft Library in the context of the 1.5 meter society, to accommodate students and staff members in their visit to the library. These topics can be formulated as either a Master thesis project or an additional thesis project.  

Project 1: Simulation-based study on social-distancing measures: design measures for the library that complies with the footsteps of the Dutch government against spreading the Covid 19, assess their performance and validity using a microscopic simulation tool (e.g., Pedestrian Dynamics) 

Project 2: Field-experiment based study on social distancing measures: test the performance of applied measures on spot, using existing sensors, manual counting, and/or survey methods 

Project 3: Field-experiment based study on behavior in the 1.5m society: use existing sensors, manual counting, and/or survey methods to investigate the behavior of visitors of the library as well as the use of the available space.

Project 4: Crowd management measures: based on the existing findings on the applied measures, and possible prediction of pedestrian flow states (to be developed), come up with crowd management measures (e.g., dynamic route information panel - DRIP) to improve the flow operations and safety in the library.

Modelling the impact of shared bicycle systems on PT demand

Type of work: Surveying, data analysis, modelling 

Potential partners: HTM 

Potential supervisor: Danique Ton

In recent years a number of shared bicycle systems has been launched in the Netherlands, amongst others OV-fiets and Mobike. On short distances these systems compete with other modes of transport, such as walking, driving and public transit. The impact of the introduction of such a system on other modes of transport is undetermined. This thesis aims to provide insights into the short- and long-term effects of a shared bicycle system on the ridership and demand distribution of intra-city public transit lines. The work features a combination of data analysis, impact assessment and the modelling thereof.  

Evacuation choice analysis in buildings using Virtual Reality

Type of work: Laboratory experiment, data analysis, modelling 

Potential partners: - 

Potential supervisors: Dorine Duives, Winnie Daamen or Serge Hoogendoorn

We all hope to never experience a real evacuation. Yet, in building designs we have to account for the possibility that its inhabitants suddenly need to exit the building due to imminent dangers of fire, smoke, etc. Little knowledge exists with respect to the impact of information on people’s exit choice behavior. The aim of this project is to develop an exit choice model for the CiTG building. As part of this project the student will use a Virtual Reality simulator to determine the effect of information on people exit choice. This project entails the design of evacuation communication strategies, the implementation of these strategies in a VR-simulator and the analysis of their effectiveness by means of a laboratory study. 

Modelling cyclist route choice (Dutch)

Type of work: data analysis, choice modelling 

Potential partners: Municipality of Amsterdam, Rijkswaterstaat 

Potential supervisors: Danique Ton and Serge Hoogendoorn 

De fiets vormt een steeds belangrijkere rol binnen ons verkeersysteem. In steden is sprake van fietsfiles, de e-bike zorgt voor een toename van het fietsverkeer, en in binnensteden en bij stations vormt fiets-parkeren een steeds groter probleem. Het aantal beleidsvragen met betrekking tot de fiets neemt toe, waardoor er zowel op nationaal, regionaal en lokaal niveau een groeiende behoefte bestaat om nauwkeuriger voorspellingen te doen voor de modaliteit Fiets. De huidige verkeers- en vervoermodellen hebben moeite om de groei van het fietsverkeer te voorspellen. Ons uiteindelijke doel is om in verkeersmodellen de kwaliteit van het modelleren van de vervoerwijze Fiets op gelijke hoogte te brengen als dat van de vervoerwijzen Autobestuurder en Openbaar Vervoer. Dat wil zeggen dat bij het voorspellen van de groei van het fietsverkeer de belangrijke factoren die een rol spelen bij de keuze voor de fiets worden meegenomen en dat het model gekalibreerd en gevalideerd is op tellingen en zoveel mogelijk rekening houdt met de routekeuzes die fietsers maken. Deze afstudeerstage vormt daar een belangrijke stap in. 

Deze stage richt zich op het kwantificeren van de fietsroutekeuzefactoren. Er is reeds veel onderzoek gedaan naar routekeuze van fietsers en eerste stappen om dit in een bestaand stedelijk verkeersmodel te implementeren zijn ondernomen (bijvoorbeeld in steden als Utrecht, Den Haag en Amsterdam). Ook internationaal krijgt de fiets steeds meer aandacht in verkeersmodellen (onder andere in Kopenhagen en München).

Potential of travel behavior change of Mobility Pilots

Type of work: data analysis, choice modelling 

Potential partners: - 

Potential supervisors: Danique Ton or Dorine Duives 

The TU Delft has hosted a large pilot, coined Pilot Mobiliteit, which is aimed at getting staff and students to change their mode of transport for their home-work trips. This one-year pilot, which is split in 4 blocks of 8 weeks, allows participants to test another mode of transport free of charge. Participants can choose between testing an electrical bicycle, public transit (incl. free OV-fiets or electrical step), or working at home. During the runtime of the pilot participants have provided information regarding their person, their household and their daily routines. The question is which of these factors are determinants for their ability to change their behaviour on the short- and long-term towards a more sustainable mode of transport. The aim of the study is to develop a mode-choice model using this (and potentially) new data pertaining e-bike usage.

Passenger traffic state estimation - Estimation of flow variables metro station in Shanghai 

Type of work: data analysis, design 

Partners outside T&P: Zhibing Jiang - Tongji University 

Potential supervisors: Winnie Daamen or Yufei Yuan 

A pilot passenger flow monitoring system based on WiFi sensors has been developed and tested in one of the major stations in Shanghai metro network, whose daily demand at normal weekdays reaches 10 million passengers. This graduation project will perform state estimation research, focus on the assessment and evaluation of this pilot passenger flow monitoring framework (and its existing modules), based on the empirical data collected from this system operating in a real-life (at a metro station), and exploring the potential enhancement of the current system. The data quality and their implications have not yet been fully investigated, besides more pedestrian estimation modules can be developed based on the available data sources. The available datasets include raw WiFi data, AFC passenger counts for a period of one week. 

Using social media images for counting people in a crowd

Type of work: data analysis, impact assessment 

Partners outside T&P: None 

Potential supervisors: Vincent, Dorine, Winnie 

Crowd managers require information on the number of people present during events or peak hours at crowded locations. As not all those places have been equipped with sensors to observe the crowd, social media images could provide additional information. Recently, methods have been compared to accurately count the number of people on an image. Which method performs best depends on the characteristics of the image (whether it is a selfie, what is the viewpoint of the camera, whether people are in the background and/or in the foreground). A method to derive these characteristics automatically has recently been developed. The objective of this study is to combine this automated way to derive image characteristics with the counting methods and to investigate how the result can be used in a crowd monitoring dashboard. 

Analyses and modelling of overtaking behaviour of cyclists

Type of work: data analysis, modelling 

Partners outside T&P: - 

Potential supervisors: Winnie Daamen and Alexandra Gavriilidou

Cyclists are subject to traffic rules just like motorized vehicles but due to their agility, size and speed, as well as the lack of law enforcement, they do not always comply with them. Policy makers and infrastructure designers are interested in and could benefit from understanding the behaviour of cyclists towards rules, so that they can develop and apply appropriate design guidelines. Due to the heterogeneity in cyclist behaviour, overtaking is a frequently observed phenomenon, which affects both capacity and safety of cycle paths. To get more insights into this behaviour, video data has been collected as part of a large-scale cycling experiment. The objective of this project is to analyse the overtaking behaviour of cyclists using trajectories, and describe this behaviour with a (choice) model.