GIS Colloquium – Talks (Summer Term 2018)

Micro Diagrams: A Multi-Scale Approach for Geovisual Analysis of Categorised Point Datasets

Mathias Gröbe
Mon, April 23, 2.15 pm (Venue: INF 348, Room 015)

Location-based social media from different platforms such as Twitter and Flickr increasingly serve as data source for many diverse research projects with their point-geocoded content. For analyses and visualisation, it is necessary to show distributions of categories in different scales and resolutions. The Micro Diagrams were developed as solution to map such large geospatial point datasets. For example, a pie chart shows the numerical proportion, and the size or transparency of the chart symbolises the number of records. Therefor an aggregation is necessary to create the diagrams and to map the number of values in one cluster to a visual variable like size. Depending on the aggregation type, the resulting patterns differ. It is possible to choose a convenient method that allows to work with multiple scales with a separate content zoom interaction and to carry out scale-dependent pattern analysis of multivariate point datasets. As visualisation constraint, the area that is used for the representation of the values scales with the numbers of aggregated values.

Voxel-based change analysis of hypertemporal terrestrial laser scanning point clouds of the research campus ARENA2036 including factory interior for the development of a digital twin

Evelyn Schmitz
Mon, April 30, 2.15 pm (Venue: INF 348, Room 015)

Raw point clouds are unstructured aggregations of measured xyz-coordinates. While it is easy for humans to detect and interpret parts of these aggregations, algorithms struggle with the correct detection and identification of objects. A solution is the selection of information given by the points of hypertemporal data acquisitions to enlarge the smartness of point clouds and their application fields, using the advantages of a fourth dimension, i.e. time. The digital shadow aims at the creation of a permanent, actual and accurate image of relevant data in the field of production. Simple and fast algorithms in combination with data from the past and present are required to develop agile, self-optimizing factories, decreasing costs by increasing production efficiency at high quality. An approach for the identification of the changed geo-location of single objects, e.g. robots, racks and trolleys, in a fabric coordinate system between several time steps is presented. Hypertemporal data were acquired at the research campus ARENA2036 with a FARO Focus S 350 laser scanner. After the application of an automated rough filtering of stable objects, a voxel-based change detection algorithm compares properties of corresponding voxels based on their constituting points. The method delivers a first step for the realization of the digital shadow in a factory hall.

Recent Advances in Remote Sensing Image Search and Retrieval from Large Archives

Prof. Dr. Begum Demir
Mon, May 7, 2.15 pm (Venue: INF 348, Room 015)

During the last decade, a huge number of earth observation (EO) satellites with optical and Synthetic Aperture Radar sensors onboard have been launched and advances in satellite systems have increased the amount and variety of EO data. This has led to massive EO data archives with huge amount of remote sensing (RS) images, from which retrieving useful information is challenging. In view of that, content based image retrieval (CBIR) has attracted great attention in the RS community. In this talk, a general overview on scientific and practical problems related to RS image characterization, indexing and search from massive archives will be initially discussed. Then, recent developments that can overcome the considered problems will be introduced by focusing on semantic-sensitive hashing based scalable and accurate RS CBIR systems.

Near-continuous monitoring of rockfalls and insights into post-seismic landslide patterns

Dr. Jack Williams
Mon, May 28, 2.15 pm (Venue: INF 348, Room 015)

This talk focuses on two strands of research that are both partially underpinned by the importance of using monitoring strategies tailored to the geomorphic change that is under examination. The first section focuses on the improved understanding of rockfall occurrence gained from near-continuous (c. 1 h) LiDAR monitoring of an actively failing coastal rockslope. Current understanding of the nature of rockfall and their controls stems from the capabilities of slope monitoring. These capabilities are fundamentally limited by the frequency and resolution of data that can be captured. An overview of the workflow and, in particular, the practicalities of 4D monitoring is provided. Monitoring at this resolution captures the importance of small rockfalls that ordinarily fail to be discretised due to their superimposition and coalescence, which has important implications for our understanding of the underlying failure mechanisms. Insights into the influence of sub-aerial drivers and the presence of accelerated deformation prior to failure are also presented. The second part of the talk focuses on patterns of landsliding in the years after an earthquake, here the 2015 Gorkha earthquake in Nepal. In addition to triggering ~25,000 coseismic landslides, the earthquake resulted in extensive and pervasive cracking on many hillslopes that did not undergo full coseismic collapse. Monitoring both new and existing landslides is critical for understanding rates of sediment mobilisation, the role of coseismic damage accumulation in driving post-seismic slope failure, and the evolving nature, extent, and severity of landslide risk. Here, initial results are presented from both ground-based monitoring and mapping from medium-resolution satellite imagery. Topographic distributions of new and developing landslides from 2014-2017 are drawn upon to suggest that a return to pre-earthquake landsliding is ongoing.

A low-cost mini-UAV laser scanning system-Kylin Cloud: design and performance

Prof. Bisheng Yang
Mon, June 11, 2.15 pm (Venue: INF 348, Room 015)

Mini-UAV laser scanning systems are receiving attractive attention for high-resolution earth observation applications. However, a compromise has to be determined between the costs, weights, and qualities of sensors because of limited payload and battery consumption of a mini-UAV (e.g., maximum payload < 5kg). Hence, a dilemma occurs to the price, accuracy, and weight of an IMU. To obtain a high quality and low-cost mini-UAV laser scanning system, this talk elaborates the design and performance a low cost mini-UAV laser scanning system—Kylin Cloud, consisting of cost-effective sensors: a MEMS-based IMU, a global shutter camera with wide angle lens, a 16-lines laser scanner, and a DJ MD 600 multi rotor UAV. On the one hand, the methods about the accurate states estimation of Kylin Cloud system and automated self-calibration of laser-IMU-camera are reported. On the other hand, the application studies of Kylin Cloud system are presented to evaluate its performance (e.g., the qualities of point clouds), showing a powerful means for typical applications, such as forest 3D mapping, power line corridor 3D mapping, and so on.

Classification of 3D Point Clouds using Deep Neural Networks

Lukas Winiwarter
Mon, June 18, 2.15 pm (Venue: INF 348, Room 015)

Per-point classification (semantic labeling) is an important step in processing topographic 3D point clouds. Current methods often rely on hand-crafted attributes to describe local point neighbourhood relations, feeding the resulting feature vectors to a state-of-the-art classifier. For classification tasks, deep neural networks (DNNs) have recently outperformed most traditional approaches. Since point clouds are unordered and irregular sets of tuples in space, the use of DNNs on point clouds has mostly been limited to strongly regularized (i.e. voxelized or rasterized) point cloud representations and their attributes. A novel method developed by Qi et al. (2017) allows the direct input of point clouds to a DNN. A feature describing a subset of points is hereby calculated using a commutative aggregation function. The commutative property solves issues with unordered input to DNNs. This method is adapted for topographic Airborne Laser Scanning (ALS) point clouds. While the neighbourhood definitions (on different scales) are required as input, the individual features describing the local point neighbourhoods are learned by the DNN in the training phase. The trained network is further evaluated in terms of robustness w.r.t. point density, distribution pattern and penetration rate. Also, attributes known to aid in classification (e.g. principal component analysis, echo broadening) can be added to see if the network further profits from this information or if the information is already learned inherently. First tests based on ALS data from the federal district of Vorarlberg (Austria) yielded an overall accuracy of 76.9%. In forested areas, this accuracy increases up to 98.6 %. Further improvements of the DNN are work-in-progress and are expected to overcome the sub-optimal classification of buildings, which are often misclassified as vegetation, as well as improve overall classification.

Big spatiotemporal data analysis based on social sensing

Jiangya Gong
Mon, July 9, 2.15 pm (Venue: INF 348, Room 015)

With the rapid development of information and communications technology (ICT), ubiquitous social sensing data bring new opportunities for us to understand our socioeconomic environments. We use the term social sensing to represent the study of characteristics of human spatio-temporal behavior, and the discovery of socio-economic environments, by various means of social sensing. This talk introduces the concept of social sensing and social sensing techniques, and discusses data, the associated analysis methods, and applications. Social sensing has brought us massive amounts of spatial data related to humans. The spatio-temporal analysis of social sensing data is working for different aspects of humans' lives, such as environment, emergency, economy, and urban planning. In the coming big data era, GIScientists should investigate theories in using social sensing data, and develop new methodologies to understand human activity based on social sensing.

Open Source Foundations for Spatial Decisions Support Systems

Prof. Jochen Albrecht
Mon, July 16, 2.15 pm (Venue: INF 348, Room 015)

Spatial Decision Support Systems (SDSS) were a hot topic in the 1990s, when researchers tried to embue GIS with additional decision support features. Successful practical developments such as HAZUS or CommunityViz have since been built based on commercial desktop software and without much heed for theory other than what underlies their process models. Others, like UrbanSim, have been completely overhauled twice but without much external scrutiny. Both, the practical and theoretical foundations of decision support systems have developed considerably over the past 20 years and I will present an overview of these developments and then take a look at what corresponding tools have been developed by the open source communities. In stark contrast to the abundance of OpenGeo software, there is currently no open source SDSS. The presentation will therefore conclude with a discussion of different approaches that lend themselves to be used as platforms for us to develop an open source framework to build an SDSS according to our needs.