Sester - Research Projects

Big Data und Machine Learning

  • TransMIT - Resource-optimised transformation of combined and separate drainage systems in existing quarters with high settlement pressure
    Increasing heavy rainfall events and growing urban districts pose great challenges for urban drainage. Using three neighbourhoods in the cities of Braunschweig, Hanover and Hildesheim as examples, it will be shown how urban development and water management aspects can be linked in the long term in neighbourhood planning.
    Leaders: Dr.-Ing. M. Beier; Prof. S. Köster, ISAH; Prof. Sester, ikg
    Team: Yu Feng, Udo Feuerhake
    Year: 2019
  • Object detection in airborne laser scanning (ALS) data using deep learning
    In partnership with the Lower Saxony State Office for Preservation of Historic Monuments, we are developing a method for automatically detecting archaeological objects in airborne laser scanning data. The type of objects to be detected are mainly those of interest by archaeologists such as heaps, shafts, charcoal piles, pits, barrows, bomb craters, hollow ways, etc. They could be point, linear, or areal objects. To this end, we are using deep learning techniques; namely, convolutional neural networks (CNNs) to classify height images from the region of interest. A combination of multiple (in most cases 5) CNN classifiers are then used to detect and localize objects of interest in a digital terrain model acquired from the region of interest.
    Team: Kazimi, Thiemann, Sester
    Year: 2018
    Sponsors: MWK Pro*Niedersachsen
  • Interdisciplinary Center for Applied Machine Learning - ICAML
    The ICAML (Interdisciplinary Center for Applied Machine Learning) aims at increasing the accessibility of machine learning across disciplines. Therefore, three fundamental components are developed and used.
    Team: Artem Leichter
    Year: 2018
    Sponsors: Federal Ministry of Education and Research
    Lifespan: 11/2017-11/2019
  • RainCars
    This idea would be easily technically feasible if the cars are provided with GPS and a small memory chip for recording the coordinates, car speed and wiper frequency. This initial research will explore theoretically the benefits of such an approach. For that valid relationships between wiper speed and rainfall rate (W-R relationship) are assumed and derived from laboratory and field experiments. Different traffic models are developed to generate motorcars on roads in a river basin. Radar data are used as reference truth rainfall fields. Rainfall from these fields is sampled from the conventional rain gauge and dynamic car networks. Areal rainfall is calculated from these networks for different scales using geostatistical interpolation methods and compared against truth radar data. The car sensors can be considered as a geosensor network. It allows to measure and process information locally in a decentralized way and thus has benefits with respect to scalability, which is crucial when large areas have to be covered with large amounts of measurement units.
    Team: Fitzner, Sester
    Year: 2017
  • Scene analysis - pattern recognition in person tracks
    The project deals with the automatic recognition of conspicuous movement patterns from given trajectories as (x, y, t) sequences of objects. For this purpose, patterns are defined that indicate conspicuous behavior for which automatic extraction methods are to be researched. This includes, among others, the qualitative description of the recognition rates of patterns.
    Team: Fischer, Sester
    Year: 2017
  • Real Time Prediction of Pluvial Floods and Induced Water Contamination in Urban Areas (EVUS)
    This project aims at developing a fast forecast model for pluvial floods in the city of Hannover. The main goal of the subproject for ikg is to integrate new sensors for the flood prediction models.
    Team: Feng, Sester
    Year: 2017
    Sponsors: BMBF Georisiken

Laser Scanning

  • TransMIT - Resource-optimised transformation of combined and separate drainage systems in existing quarters with high settlement pressure
    Increasing heavy rainfall events and growing urban districts pose great challenges for urban drainage. Using three neighbourhoods in the cities of Braunschweig, Hanover and Hildesheim as examples, it will be shown how urban development and water management aspects can be linked in the long term in neighbourhood planning.
    Leaders: Dr.-Ing. M. Beier; Prof. S. Köster, ISAH; Prof. Sester, ikg
    Team: Yu Feng, Udo Feuerhake
    Year: 2019
  • Object detection in airborne laser scanning (ALS) data using deep learning
    In partnership with the Lower Saxony State Office for Preservation of Historic Monuments, we are developing a method for automatically detecting archaeological objects in airborne laser scanning data. The type of objects to be detected are mainly those of interest by archaeologists such as heaps, shafts, charcoal piles, pits, barrows, bomb craters, hollow ways, etc. They could be point, linear, or areal objects. To this end, we are using deep learning techniques; namely, convolutional neural networks (CNNs) to classify height images from the region of interest. A combination of multiple (in most cases 5) CNN classifiers are then used to detect and localize objects of interest in a digital terrain model acquired from the region of interest.
    Team: Kazimi, Thiemann, Sester
    Year: 2018
    Sponsors: MWK Pro*Niedersachsen

Mobility

  • Traffic Regulator Detection and Identification from Crowdsourced Data
    Mapping with surveying equipment is a time-consuming and cost-intensive procedure that makes the frequent map updating unaffordable. In the last few years, much research has focused on eliminating such problems by counting on crowdsourced data, such as GPS traces. An important source of information in maps, especially under the consideration of forthcoming self-driving vehicles, is the traffic regulators. This information is largely lacking in maps like OpenstreetMap (OSM) and this research is motivated by this fact.
    Team: Zourlidou, Sester
    Year: 2020
  • Visual communication to control route choice behavior
    The individual choice of transport modality and route depends on a number of factors. In particular, information about the expected traffic situation is considered important. It should therefore be examined whether the mediation of the current and the anticipated situation on site (including the indication of certain securities) leads to the choice of a different route or even a different modality.
    Team: Fuest, Sester
    Year: 2018
    Sponsors: DFG-Graduiertenkolleg SocialCars
  • Deep learning of user behavior in road space - particularly in shared spaces
    The project aims to investigate the behaviour of different road users in unregulated spaces, i.e. spaces open to all road users. Existing approaches are based on a given movement model, which describes the individual behaviour as well as the interactive behaviour of different road users.
    Team: Cheng, Sester
    Year: 2018
    Sponsors: DFG-Graduiertenkolleg SocialCars

Bachelor Theses

  • Erzeugen von Routen unter Berücksichtigung der Sonneneinstrahlung
    Wie bereits seit 200 Jahren bekannt ist, leiden Bewohner der nördlichen Breiten überdurchschnittlich häufig unter einem Vitamin D Mangel. Um diesem entgegenzuwirken sollte sich der Mensch dem richtigen Maß an Sonnenstrahlung aussetzen. Für Stadtbewohner ist dieses Ziel während der Arbeitswoche jedoch schwer zu erreichen, da die meiste Zeit des Tages in geschlossenen Räumen verbracht wird. Gelöst werden soll dieses Problem durch ein Routing, welches Fußwege hinsichtlich der Sonneneinstrahlung optimiert. Wichtigste Faktoren sind neben der draußen verbrachten Zeit und der Fläche der von der Sonne bestrahlten Haut die Verschattung durch Gebäude und Vegetation sowie die Orientierung zur Sonne.
    Leaders: Monika Sester, Oskar Wage
    Team: Freya Wittkugel
    Year: 2020
  • Transformation of point clouds using Generative Adversarial Networks
    The goal of this thesis is to transform a DIM point cloud this way that it behaves like an ALS point cloud for the following processing steps. Firstly, both point clouds are rasterized, where each raster cell describes the distribution in height for all points within a raster cell. These rasterized images from both point clouds are then used to train a Generative Adversarial Network (GAN) such as the pix2pix network. The network outputs transformed height distributions, which can be back-projected to the original point clouds. Finally, those transformed point clouds can then be tested on different processing steps such as registration, change detection or classification.
    Team: Politz, Sester
    Year: 2019

Master Theses

  • Trajectory anomaly detection using spectral clustering and RNN-based auto-encoder
    Anomaly detection is important, because anomalous behavior may indicate critical events or objects within diverse research areas and application domains. One of such domains is transport, especially integrated urban mobility. Trajectories of moving objects are good representations of their behaviors in surveillance data and useful in detecting anomalous behavior. On one hand, trajectories can provide more agent-based, long-term information comparing with simple physical features. On the other hand, comparing with raw video data, which is usually represented as a sequence of images, trajectory data requires less storage space and computational resources. Moreover, it has a wide variety of sources, such as GPS instruments and laser-scanners.
    Leaders: Sester, Koetsier
    Team: Yao Li
    Year: 2020
  • Development of environmentally-balaced and congestion-avoiding routing algorithms by means of traffic simulation
    Due to the constantly growing volume of traffic in urban environments and the resulting problems such as increased air pollution, environmentally oriented approaches to achieve better urban sustainability of transport play an increasingly important role. This thesis deals with the development of environmentally-friendly routing algorithms and their validation in traffic simulations. The routing algorithm used is the A* - algorithm using the developed criteria as weights.
    Leaders: Sester, Fuest
    Team: Christian Hartberger
    Year: 2019
  • Multi-Path Prediction of Mixed Traffic Trajectories in Shared Spaces
    In shared spaces, road signs, signals, and markings are removed to allow mixed traffic directly interact with each other. The traffic engineer Reid defined it as a street encouraging pedestrian movement and reducing the dominance of vehicles without explicit traffic rules. All users have to follow informal social protocols and negotiation to use the road resources, and avoid any potential collisions. The lack of regulations makes interactions between multimodal road users more complex compared with conventional designs. With the availability of large scale datasets and the development of deep learning techniques in sequence modeling and prediction, deep learning approaches are widely used for trajectory prediction.
    Leaders: Hao Cheng, Prof. Sester and Prof. Fidler
    Team: Xinlong Han
    Year: 2019
  • Scene Context-Aware Trajectory Prediction in Shared Space
    In shared spaces, road signs, signals, and markings are removed to allow mixed traffic directly interact with each other. At a micro level, understanding how they behave and how we can foresee their behavior after a short observation time are crucial to intent detection and autonomous driving, and traffic management in shared spaces.
    Leaders: Hao Cheng, Prof. Sester and Prof. Fidler
    Team: Rui Liu
    Year: 2019
  • Residual Learning for Mixed Traffic Prediction in Shared Space
    In recent years, with the increased availability of computational power and large-scale datasets, data--driving approaches, especially Deep Learning approaches, have been largely used for trajectory modeling. Nevertheless, predicting mixed traffic trajectories in shared space is not trivial.
    Leaders: Hao Cheng, Prof. Sester and Prof. Fidler
    Team: Yuhao Zhang
    Year: 2019
  • A Study of State-of-the-Art DL Methods for Mixed Traffic Trajectory Prediction
    In recent years, with the increased availability of computational power and large-scale datasets, data-driving approaches, especially Deep Learning (DL) approaches, have been largely used for trajectory modeling. The performance for pedestrian trajectory prediction in crowded spaces has been improved year by year, such as the state-of-the-art Social-LSTM (Alahi et al., 2016) CVAE (Lee et al., 2017), and Social-GAN (Gupta et al., 2018). The goal of this master thesis is to apply such stat-of-the-art DL approaches in a more challenging environment—shared space—for trajectory prediction with mixed traffic agents and compare their performance.
    Leaders: Hao Cheng, Prof. Sester and Prof. Fidler
    Team: Xin Xu
    Year: 2019
  • Transformation of point clouds using Generative Adversarial Networks
    The goal of this thesis is to transform a DIM point cloud this way that it behaves like an ALS point cloud for the following processing steps. Firstly, both point clouds are rasterized, where each raster cell describes the distribution in height for all points within a raster cell. These rasterized images from both point clouds are then used to train a Generative Adversarial Network (GAN) such as the pix2pix network. The network outputs transformed height distributions, which can be back-projected to the original point clouds. Finally, those transformed point clouds can then be tested on different processing steps such as registration, change detection or classification.
    Team: Politz, Sester
    Year: 2019
  • Automatische Parametrische Beschreibung von Bodendenkmalen
    Die Untersuchung von Bodendenkmälern liefert wichtige Erkenntnisse zur Entwicklung der Kulturlandschaft in Deutschland. Sie lässt Rückschlüsse über das Leben in vergangenen Epochen und die Veränderung zwischen den Zeitaltern zu. Die Lagebestimmung oder die messtechnische Erfassung der Bodendenkmäler ist nicht immer einfach. Oftmals lassen sich die Objekte schwer erkennen oder befinden sich an schwer zugänglichen Stellen. Airborne Laserscanning eröffnet hierbei eine vergleichsweise neue Methode der archäologischen Prospektion. Aus den erzeugten Laserscannerdaten lassen sich hochauflösende flächendeckende Geländemodelle erzeugen, mit denen auch vorher unentdeckte Bodendenkmäler in Wäldern erkannt werden können. Die auf diese Weise erzeugte Datengrundlage bietet nicht nur neue Möglichkeiten, sondern auch neue Herausforderungen. Durch die hohen Datenmengen stellt sich die Frage nach der Bearbeitungszeit und die damit verbundene hohe Arbeitszeitbelastung. Um diesem Problem zu begegnen, ist es sinnvoll möglichst viele Prozesse zu automatisieren.
    Leaders: Frank Thiemann, Monika Sester
    Team: Dennis Elschen
    Year: 2018
  • Traffic-sign Recognition from Street-level Photos: a Deep Learning Approach
    The scope of this thesis is the recognition of traffic-signs from street-level images. A state-of-the-art deep learning detection algorithm is used, the Single Shot Multi-box detector (SSD) and within the thesis its performance is validated experimentally by examining different training strategies.
    Leaders: Prof. Dr. Bernhard Roth, Co-advisor: Prof. Monika Sester, Supervisor: Stefania Zourlidou
    Team: Qifa Bao
    Year: 2018
    Lifespan: 2018