| 11.10.2022 |
Institute Meeting (ifgi staff only) |
|
|
|
Sorry, no abstract available yet. |
| 18.10.2022 |
Spatiotemporal functional models for irregular and large environmental data sets |
Prof. Dr. Philipp Otto |
Leibniz University Hannover |
|
In this talk, I will show the results of two different articles -- one about the morphological evolution of coastal profiles and one about the usage patterns of a bike-sharing system. Whereas the first data set is highly irregular and often has incomplete measurements, the second data set has a very detailed but regular temporal resolution. For both cases, we applied a functional model, which accounts for latent spatial and temporal effects, in combination with a spatial subsampling/bootstrap approach. I will show how this model can be used in two completely different situations and how we made the procedure scalable while accounting for the full spatial and temporal dependence.
In general, beach profile data sets provide valuable insight into the morphological evolution of sandy shorelines, but to be precise, our aim was (1) to determine the temporal and spatial variability of beach profiles while accounting for autoregressive dependencies, (2) to identify effects of external influences, (3) to predict complete beach profiles at unknown locations, and (4) to forecast complete beach profiles accounting for external influences, such as storm events or nourishments.
Regarding the second analysis, the station hire data is analysed in a spatiotemporal functional setting, where the number of bikes at a station is defined as a continuous function of the time of day. Understanding the usage patterns for bike-sharing systems is essential in terms of supporting and enhancing operational planning for such schemes. Studies have demonstrated how factors such as weather conditions influence the number of bikes that should be available at bike-sharing stations at certain times during the day. However, the influences of these factors usually vary over the course of a day, and if there is good temporal resolution, there could also be significant effects only for some hours/minutes (e.g., rush hours). |
| 25.10.2022 |
Geospatial technologies meet health: prediction of air pollution exposures at high spatiotemporal resolution |
Carles Milà |
Pompeu Fabra University |
|
Outdoor air pollution is the largest environmental risk factor for human health in Europe, with more than 300,000 premature deaths attributed to it every year. In order to estimate the association of the different air pollutants with health outcomes, epidemiologists require exposure estimates at high spatiotemporal resolution that can be linked to study participants. Nonetheless, air pollution monitoring is complex, expensive, and limited to a generally small number of locations. Therefore, predictive modelling approaches are frequently used to estimate spatially-continuous concentrations of the pollutants using atmospheric remote sensing, climate and atmospheric forecasts, as well as other geographical data. In this talk, I will review the challenges found in air pollution exposure assessment including ground measurements, spatial sampling, atmospheric remote sensing, and exposure modelling and validation. As an example, I will present an exposure assessment study where we modelled daily concentrations of PM2.5, PM10, NO2 and O3 for the period 2018-2020 to assess the association between air pollution and COVID-19 severity in Catalonia, Spain (COVAIR-CAT project). Briefly, a 2-step modelling framework was used where 1) missing pixels in the atmospheric remote sensing products were reconstructed via imputation models using atmospheric forecasts, and 2) a quantile random forest algorithm with residual kriging was used to model the concentrations measured at the ground stations while quantifying the model uncertainty in the predictions. Strengths and limitations of the approach and future lines of work will be highlighted. |
| 08.11.2022 |
Sustainable Development Research for Insect Monitoring & Protection - Using Computer Vision & Machine Learning Systems |
Prof. Dr. Benjamin Risse |
ifgi |
|
Over the past 50 years insects have declined by 75%.
This decline has catastrophic implications for our ecosystems and all terrestrial and freshwater food chains and food webs.
Quantifying these developments is however notoriously difficult for current computer vision and machine learning systems:
Tiny, low contrast and fast moving objects have to be localised in cluttered and dynamic scenes resulting in visual ambiguities, frequent occlusions and the need to process huge video files with inappropriate foreground-background ratios and sparse visitation events.
In my presentation I will summarise some of the most prominent challenges of visual insect monitoring and introduce solutions and tools to extract high-throughput and high-accuracy characteristics for a wide range of ecological and biological applications.
In particular, I will discuss our progress in building tools and algorithms to quantify the impact of pesticides on insect behaviour, present insights into our attempts of building a versatile insect camera trap and in developing an appearance-agnostic in-field animal detection and tracking algorithm to extract accurate measurements from camera footage.
I will close by presenting complementary environmental mapping approaches that yield rich contextual information, and moreover can be combined with behavioural data to achieve the urgently needed lab-level quantifications in wildlife conditions. |
| 15.11.2022 |
A Glimpse into Statistical Relational AI: Indistinguishability for the Win! |
Jun.-Prof. Dr. Tanya Braun |
WWU Münster - Computer Science Department |
|
Statistical relational artificial intelligence, StaRAI for short, focuses on combining reasoning in uncertain environments with reasoning about individuals and relations in those environments. An important concept in StaRAI is indistinguishability, where groups of individuals behave indistinguishably in relation to each other in an environment. This indistinguishability manifests itself in symmetries in a propositional model and can be encoded compactly using logical constructs in relational models. Lifted inference then exploits indistinguishability for efficiency gains. This talk showcases how to encode indistinguishability in models using logical constructs and highlights various ways of using indistinguishability during probabilistic inference. |
| 22.11.2022 |
Building bridges with pen and paper: Sketch maps make local knowledge visible |
Dr. Carolin Klonner |
Heidelberg University |
|
Worldwide, more and more people are affected by floods. The disaster in the Ahr valley shows us very clearly that new methods are urgently needed to obtain up-to-date and locally relevant information. This is where the inhabitants of the affected regions come into play, because their local knowledge can serve as a valuable supplement to data collected with technical devices. The question arises as to how this knowledge can be made visible and usable. As an answer to this question, Dr. Carolin Klonner from the Department of Geography at the University of Heidelberg will present the sketch map method, which can be used for river floods and flash floods. Her research areas are in Chile, Germany and Brazil. |
| 29.11.2022 |
Enhancing navigation services for pedestrians - Selected research results of the research group cartography of TU Wien |
Prof. Dr. Georg Gartner |
TU Wien |
|
Navigation services are prominent examples of Location-based Services and have become quite popular for pedestrians in indoor- and outdoor environments. In this presentation I will focus on selected aspects of enhancing pedestrian navigation services by applying a user-centred focus. Case studies demonstrate various attempts, including the use of signs as landmarks, integrating subjective relations to places and introducing a semantic framework. |
| 06.12.2022 |
Fostering walking in the city with geospatial methods – three related works |
Dr. Tessio Novack |
ifgi |
|
Walking has several physical and mental health benefits and should be incentivized by cities valuing sustainability and social integration. In this talk, I present three of my recent works related to fostering walking with geospatial methods. The first work introduces an application for generating customized pleasant pedestrian routes with data from OpenStreetMap. The second work is on detecting building facades with graffiti artwork based on user-generated photos and Google Street View imagery. The third work is an ongoing one. Based on a user-survey, Google Street View imagery, and image feature extraction, I address the questions of (1) which streetscape attributes correlate with positive and negative perceptions of street walkability and (2) how this correlation is dependent on socio-demographic characteristics. Lastly, I quickly discuss how this ongoing work links to my initial research plans at ifgi. |
| 13.12.2022 |
Lowering the barrier for modern cloud-based geospatial (big) data analysis: the Geospatial Computing Platform |
Dr. Ing. Serkan Girgin |
ITC, University of Twente |
|
Geospatial data is getting bigger and such large and complex datasets are becoming more and more difficult to process by using traditional systems and methods, such as individual workstations and single-threaded applications. Numerous spatial computing solutions have been developed to tackle this challenge by enabling distributed data stores, parallel and distributed computing capabilities, and special computing units (e.g., GPU/TPU) to enable discovery, delivery, analysis, and visualisation of geospatial data. However, these solutions require specialized know-how and expertise, as well as access to adequate computing infrastructure that is mostly located remotely in the Cloud. Therefore, a transition in modus operandi is necessary. The Geospatial Computing Platform lowers the barrier by providing a state-of-the-art computing infrastructure designed for (big) geospatial analysis tasks that combines low-energy, high-performance Edge AI units with powerful GPU-enabled big data computing units in a seamless and innovative manner. Through the platform the users can access thousands of scientific software packages (e.g. Python / R) that are kept up to date regularly. Public datasets available platform-wide improve data access and reduce data duplication, whereas shared workspaces allow research groups to work in a collaborative manner. Beside a modern interactive notebook interface, the platform also allows remote desktop access for desktop applications (e.g., QGIS, SNAP) and features integrated geospatial database, map serving, and data collection services to benefit from existing well-established tools and technologies. This talk will provide information about the design and architecture of the platform, current use cases, and lessons learned during the operation period of two years, involving 200,000+ hours of multi-core/GPU computation and a user community of more than 800 users. |
| 10.01.2023 |
[cancelled] Dimitar Valkov |
|
|
|
Sorry, no abstract available yet. |
| 17.01.2023 |
(12:15) The nexus of spatial analysis and digital geographies |
Jun.-Prof. Dr. René Westerholt |
Department of Spatial Planning, TU Dortmund University |
|
Spatial statistical analysis is an offspring of the so-called quantitative turn that affected large parts of the social sciences from the 1950s onwards. The core of spatial analysis developed at the intersection of statistics, regional science, economics, and geography, complemented by contributions from application domains such as ecology and epidemiology. In terms of analysing human-geographic phenomena, the field is conceptually underpinned by ideas that respond to a largely analogue world. In that world, human spatial experiences were hardly influenced by real-time information and there was no profound interweaving between everyday geographical experience and digital technologies. These circumstances have changed fundamentally in recent years. Digital technologies, whether visible or invisible, are now routinely embedded in everyday practices. Geographers have begun to theorise the new kinds of geographies that emerge from these socio-technical contexts, but spatial analysis has so far hardly responded to these new kinds of spaces. This talk begins with an overview of digital geographies and corresponding space concepts. Building on this, challenges for spatial analysis will be discussed, including some contributions the speaker has made in recent years. |
| 17.01.2023 |
(14:15) Liberation from the physical constraints of the service space |
Assist. Prof. Keiichi Zempo |
University of Tsukuba |
|
Physical laws do not bind interaction through VR space. Therefore, remote persons can wear avatars of their choice, easily connect with remote persons, and individually customize the shared space, for example, by adjusting the volume individually. In the first half of this presentation, we will show examples of service applications that break physical constraints in VR space, and in the second half, I will show how to create contact points in VR space without using HMDs.
First, I introduce the audiovisual experience in a virtual reality (VR) service context that enables a more effective interaction between a user immersed in a virtual environment (VE) and an avatar as a store staff. By utilizing the characteristics of VE experiences, we find the effects of this unrealistic relationship between the visual and auditory positions of the avatar presented to the user variable rather than uniformly presented in the same position. In this study, we experimented with investigating how the positional deviation between the sound and visual images can be tolerated in VE, the effect of positional deviation on the interpersonal distance to the avatar, and the possibility of manipulating the impression of the avatar by deviating the sound image from the visual image.
Next, I present research on hardware that enables interaction with virtual objects without the need to wear an HMD.
This study seeks to demonstrate that a navigation system using stereophonic sound technology effectively supports visually impaired people in public spaces. In the proposed method, the stereophonic sound is produced by a pair of parametric speakers for a person who comes to a specific position, detected by an RGB-D sensor. The sound is a stereophonic earcon representing the target facility. The recipient can intuitively understand the direction of the target facility.
K. Zempo, A. Yamazaki, N. Wakatsuki, K. Mizutani and Y. Okada, Mouth-in-the-Door: The Effect of a Sound Image of an Avatar Intruding on Personal Space That Deviates in Position From the Visual Image, in IEEE Access, vol. 10, pp. 125772-125791, 2022, doi: 10.1109/ACCESS.2022.3222804.
Y. Mashiba, R. Iwaoka, H.E. Bilal Salih, M. Kawamoto, N. Wakatsuki, K. Mizutani and K. Zempo, Spot-Presentation of Stereophonic Earcons to Assist Navigation for the Visually Impaired, in Multimodal Technol. Interact. vol. 4, 42, 2020, doi: 10.3390/mti4030042.
N. Kuratomo, H. Miyakawa, S. Masuko, T. Yamanaka and K. Zempo, Effects of acoustic comfort and advertisement recallability on digital signage with on-demand pinpoint audio system,
In Applied Acoustics, vol. 184, 108359, 2021, doi: 10.1016/j.apacoust.2021.108359. |
| 24.01.2023 |
[ERCIS Lunchtime Seminar] New innovation paradigms for networked urban ecosystems |
Rui José |
Universidade de Minho |
|
Becoming a smart city is now a common aspiration for cities across the world. Many have established innovation ecosystems to promote the digital transformation of their territories, embracing a broad range of dimensions, such as mobility, energy, digital governance, sustainability or the entrepreneurial context. A few years into this transformation process, we now have a considerable body of research literature reporting on the successes and limitations of various types of innovation strategies and how they have been applied across numerous smart city initiatives. The promotion of local innovation ecosystems composed of multiple types of stakeholders, including citizens, is often mentioned as a key success factor. However, this idea of many independent ecosystems, each focused only on a specific territory, can also lead to considerable fragmentation and may fail to offer the type of combinatorial innovation and shared learning that is essential for the consolidation of successful practices and for the acceleration of the transformation process. This talk analyses how the promotion of extra-ecosystems connections may unlock new sources of innovation and discusses some of the areas where this approach could be more valuable.
Short Bio: Rui José got his PhD in Computer Science in 2001 from Lancaster University, UK, and he is now a full member of Centro Algoritmi, where he leads the Urban Computing Lab. His research interests include Ubiquitous Systems, Sustainable Mobility, Smart Cities and Digital Innovation. He has over 140 publications in books, book chapters, journals and conference proceedings. Rui José has been PI or co-PI in 9 competitive funding projects, including projects funded by international programs, such as FP7, FET-OPEN and Portugal/CMU. Rui José is also an entrepreneur, having created 3 companies, two of which were officially awarded the label of University of Minho spin-offs. |
| 31.01.2023 |
Institute Meeting (ifgi staff only) |
|
|
|
Sorry, no abstract available yet. |
