GIS Colloquium – Talks (Winter Term 2017/2018)
Mon, October 23, 2.15 pm (Venue: INF 348, Room 015)
Digital 3D city models are of crucial importance in many applications such as urban and regional planning and enable in the environmental field precise analyses and simulations of pollutant, flood, and noise propagation. Their manual reconstruction provides good results, but is usually very time-consuming and expensive. In order to overcome this issue, the development of automatic reconstruction approaches for the time-efficient and cost-effective generation of 3D building models has become of great interest in recent years. In this talk, a fully automatic building reconstruction approach will be presented which uses building points of an aerial LiDAR data set. The approach is characterized by a strong integration of building knowledge, which is automatically derived during the reconstruction through the application of a graph grammar. It utilizes half-space modeling techniques for the construction of 3D building models to ensure their topological correctness. The resulting building models feature many details and provide in addition to the geometric information also semantic information if required. Thus, they are well suited for different applications. The talk will conclude with a brief overview of related research activities of the speaker.
Mon, November 27, 2.15 pm (Venue: INF 348, Room 015)
The speaker tries to give a structured and comprehensive overview of event detection from geo-tagged twitter data and present some open questions. Precisely, it starts with an introduction of the basic conceptions to answer questions like: what is event and event detection in social media, then an overall workflow of event detection will be given. Based on existed techniques applied in recent papers, three different approaches of event detection focus on: clustering driven approaches, anomaly detection driven approaches, and topic modelling driven approaches will be analysed in terms of their algorithms and advantages and disadvantages. The talk ends with some open questions regarding the research challenges (e.g. scale problem), and trends (e.g. multi-source event detection) in the field of event detection from the perspective of GIScience.
Dr. Alexandra Diehl
Mon, December 11, 2.15 pm (Venue: INF 348, Room 015)
In this talk, I will introduce my work in the area of Visual Analytics and Predictive Analytics with applications in Weather Forecasting and, more recently, in Social Media. When Predictive Analytics reaches its limits, Visual Analytics can help to recalibrate, change, and optimize models, and to validate results. Putting the user in the loop – the main goal of Visual Analytics – allows the analysts to inspect internal parts of predictive algorithms, such as regression models, and optimize them for a better fitting. I will present several design studies and contributions to this research area in the form of Multiple Coordinated View System and Visual Analytics workflows. My main goal for the ongoing and future projects is to overcome the limitations of automated Predictive Models using the experience and knowledge of the user.
Travel History: Reconstructing Travelers Semantic Trajectories Based on Heterogeneous Social Footprints
Amon Veiga Santana
Mon, December 18, 2.15 pm (Venue: INF 348, Room 015)
Travel specialized services on the web have increased their sociability and usage by adopting mechanisms that facilitates content sharing in real time between users. These web applications, however, lack tools that allow travelers to share their experiences, such as places they have visited, itineraries they have performed, and other activities of a typical touristic trip. These inds of information, when available, are insufficient and incomplete. The process of generating structured and semantic rich datasets based on recommended trips, routes and destinations usually requires high effort to be generated. This task is frequently manual, cumbersome, inaccurate, time-consuming, and depends on user’s willingness to cooperate. This work proposes a solution for reconstructing travel histories using heterogeneous social sources, such as posts in social networks, GPS positioning data, location history data generated by cloud services or any digital footprint with an associated geographic position. The solution encompasses a conceptual model; a methodology to reconstruct travel histories based on heterogeneous social tracks sources; and an application to present the reconstructed travel itinerary in a graphical and interactive fashion. An experiment conducted with real travelers showed that the proposed solution is a reasonable way to reconstruct semantic-rich travel histories in an automatic fashion.
Dr. Mohammad Sharif
Mon, January 22, 2.15 pm (Venue: INF 348, Room 015)
Studying movement in geographic information science (GIScience) has received attention in recent years because it plays a crucial role in understanding and modeling various spatial activities and processes. In reality, movement of an object is embedded in context and is highly affected by both internal and external contexts. The former is any factor that is related to the object’s characteristic, state, and condition, while the latter is dedicated to the environmental conditions during the move. Such consequential influence has created new paradigms for context-aware movement data mining and analysis. Among the potential movement analysis research, studying moving point objects (MPOs) and measuring the similarities between their trajectories have been of interest recently because it can be the basis for understanding objects’ behaviors, extracting their movement patterns, and predicting their future movement trends. Despite such importance, less attention has been paid to contextualizing similarity search of trajectories, so far. In this research, after providing a new definition and a taxonomy for context in movement analysis, a series of distance functions have been developed for assessing the similarities of trajectories, by including not only the spatial footprints of MPOs but also a notion of their internal and external contexts. In other words, the degree of similarity between two trajectories not only is related to the spatial and temporal closeness of trajectories but also is highly associated with the commonalities in the contexts that they share. The effectiveness of the developed methods have been examined in several experiments on real datasets, i.e., commercial airplanes’, pedestrians’, and cyclists’ trajectories, in separate study areas, while accounting the internal and external context information during the movement. The results of these implementations demonstrate the significance of incorporating contextual information in movement studies, as movement is highly affected by context in both positive and negative manners.