This PDF file contains the front matter associated with SPIE Proceedings Volume 12269, including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

We present the concept of a mobile measurement platform paired with an end-to-end data processing chain that enables analysis of multimodal sensor data in real time for smart city mapping. The proposed system can be integrated on mobile platforms into everyday traffic in urban environments. The online pre-processed and compressed information can then be used to directly update a cloud-based digital twin. This enables the creation of a virtual image of entire cities and generates data that can be used for real time Urban Information Modeling, and thus a valuable planning tool to provide up-to-date information at any time. The generated data form the basis for decision-making on improving mobility flows for smart transportation systems and autonomous vehicles and the survey of infrastructure and vegetation for sustainable urban development. The proposed concept is achieved using energy efficient embedded sensors and processing units in combination with computational optimized software architectures close to the sensors.

The automated segmentation of photogrammetric or LiDAR PC data offers the basis for further post-processing. One example is the use of enriched data for simulating wind, heat, and rain to increase the resilience of cities against stronger and more frequent weather events due to climate change. The semantic information can be used to distinguish tree species or surface types within simulations for more precise results. Apart from that, many other applications like autonomous driving or disaster management can benefit. However, the quality of the segmentation highly depends on the amount and quality of annotated training data and the applied segmentation method. This paper gives an overview and compares state-of-the-art PC segmentation methods including both 2D, 3D and combined approaches. Furthermore, current annotation tools and methods are evaluated regarding the accuracy, the number of classes, and time effort. Moreover, current strategies and methods to extend training data with augmented or synthetic data are listed and assessed. Besides, we present a concept for a specific segmentation task based on a trade-off of the previously listed methods. The task includes the segmentation and partial annotation of colourized PC data of two urban areas in Freiburg Germany, acquired from a driven and flying platform. Lastly, we present the required steps to extend the segmentation task from one city section to an overall city with diverse areas and to adapt to different sensor systems leading to different PC specifications. We conclude this paper by identifying challenges and required research in the field of PC segmentation and annotation.

An effective maintenance plan of railways bridges needs suitable, user-friendly tools for preventive inspection. The latter, actually, when applied to large infrastructures such as railways bridges, requires a large amount of resources and it is time consuming. Presently, most inspection relies only on visual inspection, often with a lack of adequate procedures for documentation filing and an easy consultation of it. In this paper we present a series of experiments conducted with the aim of developing an ICT tool for supporting the inspection of railways bridges. The developed tool integrates a full range of diverse data and guarantees easy access for their storage and consultation. The data integrated in the ICT platform includes images acquired with diverse techniques (3D scanning, high definition photography, photogrammetry, thermography, fluorescence LIDAR) and integrates them with data from traditional survey methods. In order to make easier the consultation to the user, the data are integrated onto a 3D model of the bridge. In particular, we will focus on the results obtained by using the fluorescence LIDAR and 3D LIDAR scanning technique on a masonry bridge for its preventive inspection and their integration in the ICT platform. All data were also complemented with the results from traditional survey methods. False-color coded thematic maps based on fluorescence LIDAR data highlighted the presence of biofilms and patinas on the bridge’s surface, providing complementary information with respect to those from thermographic images at large.

Due to the relatively high average age of the rail infrastructure in Germany and the thus often historic plans, as-built documentation has a very high priority at Deutsche Bahn AG. The inventory and updating of existing plans represent an enormous challenge for the operator, DB Netz AG. More than 4.6 million inventory plans must be continuously checked to ensure that they are up to date, correct, adjusted and supplemented as necessary. The most fragile structures are railroad bridges. These are the focus of this paper. For now, all information of bridges such as planning documents, statics, status reports of bridge examination, etc. are collected in decentral locations of the owner or operator. The existing information is available in a wide variety of formats, e.g. pdf files, plans on paper, scanned paper plans, digitally created plans, SAPdata and photos. We tackled this problem of non-uniform and decentralized data management within the mdfBIM project. Within the scope of this project, a process model was developed that describes the merging of the various data sources in the planning process and attempts to identify the primary data source in each case. The validation and adaptation of this model was carried out continuously after it had been set up based on a railway bridge in Hannover, Germany. We used machine learning algorithms to enable an automated object classification for the most common objects to derive the highest possible degree of automation. Another important step towards automation was the consolidation of the numerous data sources. This existing, inhomogeneous data was homogenized in a defined process. During this homogenization, the data sets - ranging from existing as-built plans, photo documentation, maintenance and conversion reports, SAP extracts, construction books and construction plans to the newly recorded laser point cloud - was evaluated.

In this paper; the complete process chain and the first results are presented. Furthermore, an outlook is given on further research tasks and the further development of the elaborated process chain.

This paper aims to analyze the evolution of land-use patterns and the Eco-Environmental Quality Index (EQI) in major European metropolitan areas. To this end, nighttime remote sensing images will be used to delineate ten major European metropolitan areas with more than 1 million inhabitants and CLC land cover data will be used to reclassify European land occupation types from 2000 to 2018. Based on MODIS satellite remote sensing imagery, the evolution of EQI was studied for the EQI evaluation systems applicable to different periods and different regions. The results show that in Europe, cultivated cover and discontinuous built-up cover occupy the largest part of the land, but the former is shrinking and the latter is expanding. Other land areas are much smaller than the previous two. Among them, grassland has always shown a growing trend, and the fastest growing area is industrial land, while the slow-expanding land includes transportation land, leisure land and water bodies. The overall result of the EQI was estimated to be around 0.59. The average result in 2000 was the highest, with a decrease in 2006, but a small increase since then. The ecological environment of each metropolitan area has experienced both improvement and deterioration. Overall conditions improved in Barcelona, the English central agglomeration, Rome and the diffuse landscape of Northern Italy, London and Naples. But The Belgian/Dutch agglomeration, the lands of Paris and the Rhine-Ruhr, the vast majority of the ecological quality is deteriorating. Madrid's EQI worsened the most.

Mobile mapping vehicles are rapidly gaining in importance as they accelerate the digitization of road surroundings. This data is indispensable for municipalities as a quantification of the state of the art as well as for planning of infrastructure projects. Fast laser scanners for 3D point cloud generation and high-resolution cameras are becoming more readily available, increasing not only data quality but also the amount of acquired raw data. Thus, taking advantage of modern computing power and developing strategies for data handling is getting crucial. In this paper, a measurement vehicle is presented that handles up to 4 GB/s of raw data. The vehicle is based on a modular system platform from the Fraunhofer Institute for Physical Measurement Techniques IPM. A Clearance Profile Scanner built by Fraunhofer IPM scans a 345° area almost perpendicular to the driving direction and delivers 3D data with 2 MHz sampling rate and millimeter precision. A high-end positioning solution provides the absolute location at any time. Two off-the-shelf panoramic cameras provide images for standard use cases. As a special feature, six 31 megapixels CMOS cameras acquire high-resolution image data from which a panoramic image can also be computed. A special network controller running custom software handles the large amount of network traffic. Additionally, a sophisticated on-board data processing pipeline was developed which performs image compression on a graphic card before the data is stored in permanent memory. Thus, all raw data can be managed and stored for driving speeds up to 144 km/h.

Unmanned Aerial Vehicles (UAV) provide increased access to unique types of urban imagery traditionally not available. Advanced machine learning and computer vision techniques when applied to UAV RGB image data can be used for automated extraction of building asset information and if applied to UAV thermal imagery data can detect potential thermal anomalies. In this work, we present a machine learning approach for asset extraction of a building’s roof top unit (RTU) using a state-of-the-art object detection algorithm. We also present an approach to identify potential thermal anomalies on the building envelope. Our object detection algorithm achieves 89% accuracy on the test dataset, while our thermal anomaly algorithms are able to identify potential anomalies, but require further testing for accuracy. The asset information and anomalies are relevant to a variety of urban and energy applications.

Industrial areas identification is an important problem in detecting urban land use. The industrial area contributes significantly to the carbon footprint and economic status of the region. Industrial buildings/factories are marked by metal roofs. We attempt to leverage this reasonably characteristic association for detecting industrial buildings. The foundation of our study is the spectral properties of industrial roofs which have high reflectance and flat spectrum. With these spectral properties of metal roofs, we have designed an algorithm with less time complexity as compared to the other approaches like matching reference signature with every pixel in the image or matched filter target detection approaches. The algorithm to detect the metal roof and hence industrial shade is divided into two main parts: 1. Calculating the relative reflectance of the image. 2. Calculating the spectral flatness of the pixels. In step one we use the high reflectance characteristics to calculate the relative reflectance of the image based on percentile brightness. In step two we use the flatness of the spectrum with the mean of consecutive band ratios. Thresholding on this band ratio gives us the industrial roof pixels. The algorithm is tested on the very well known hyperspectral images like Pavia University and Urban Image.

The variation of air temperature (T_air) over a spatial area is usually hard to detect. This is because it requires a high-spatial network of weather stations distributed over the study area which might be impractical to implement in large studies. Nevertheless, remotely sensed images could provide a low-scale land surface temperature (LST) map for large areas where one of the renowned satellites that provide LST is Landsat. This research tries to understand the accuracy of these LST maps using records from weather stations as well as thermal camera images. To do so, remotely sensed images between April 2021 and June 2022 were retrieved from Landsat 8 to estimate the LST over the University City of Sharjah in the UAE to validate them using 14 weather stations distributed on different land uses. The weather stations recorded T_air every 30 minutes at 0.5 meters and 1.5 meters above the ground. In addition, thermal images from the thermal camera were taken at the same time as Landsat 8. The correlation analysis shows that the relationship between LST and the thermal camera is weaker during higher temperatures, while strong at lower temperatures (R² of 0.878) with an absolute difference of about 8°C. On the other hand, the relationship between LST and T_air at 0.5 meters is slightly stronger than the one at 1.5 meters (R² of 0.877) and (R² of 0.870) respectively.

Semantic segmentation of urban areas can provide useful information for analyzing and detecting changes in urban development. Recently, numerous remote sensing image datasets from various platforms have been acquired, and various semantic segmentation studies using them have been conducted. However, they do not contain many images because of their large data capacity and difficulty in constructing label data. Furthermore, it is difficult to use them simultaneously because each dataset has a different spatial resolution, shooting angle, and meaningful objects. In this study, two different UAV image datasets, such as UAVid semantic segmentation and semantic drone datasets, were used to train a combined U-net model to use heterogeneous remote sensing datasets for semantic segmentation tasks simultaneously. The UAVid dataset has a flight height of 50 m and 300 images with eight classes. However, the semantic drone dataset was acquired at an altitude of 5–30 m above the ground and contains 598 images with 20 classes. The combined U-net model is based on the U-net architecture, but it receives input from two different data sources. The experimental results showed that learning two datasets with a combined U-net improved semantic segmentation accuracy more than learning each data with a U-net. This study confirms the ability to train two different datasets acquired from different places and platforms simultaneously; thus, evaluating the applicability of semantic segmentation studies using heterogeneous remote sensing datasets.

The paper describes an approach to restoring a three-dimensional model of rigid objects from a single satellite image based on informative classes identified from the results of machine learning, which include railway rails and poles, roofs and walls of buildings, shadows of poles and buildings, and others. The proposed algorithms take into account various conditions for the presence of certain classes in the image, identified by the results of machine learning, as well as the conditions for the absence of metadata on the spatial resolution and spatial orientation of the shooting and the Sun (shooting angle, scanning azimuth, etc.).

We describe the digital technological platform SODIS Building CM for visual control of construction using a BIM model. It can track the work schedule and control the processes of delivery and acceptance of construction and installation works. BIM model elements can be associated with various tasks, documents, equipment, and processes. Elements of the BIM model can automatically change color depending on the state of the object. Staining changes depending on the percentage of completion and the backlog from the base schedule of work. Using a 3D model of an object allows control the amount of work performed on structural elements and all engineering systems. The article also discusses the capabilities of the SODIS Building FM platform, which allows you to get, instead of a pile of operational documentation, which is extremely inconvenient to use, an almost ideal system for safe operation, where the search for the necessary information takes a matter of seconds. This is especially important in the event of an emergency, when there is an acute shortage of time to make effective management decisions.

Zhengzhou City suffered the strongest rainstorm in history in July 2021, floods and other secondary disasters have greatly endangered the property and safety of local residents in Henan Province, China. In this paper, the flood risk of Henan Province was analyzed from the three perspectives of hazard, exposure and vulnerability, and seven factors of rainfall, elevation, slope, distribution density of river network, population, GDP and arable land occupation ratio were selected to create a flood disaster risk evaluation index system. Remotely sensed data and GIS were used to analyze and map the influencing factors, and then the comprehensive risk results of flood disasters in Henan Province were obtained based on analytic hierarchy process (AHP) method. Results show that the higher-risk area and high-risk area mainly locate in cities of Xinyang, Luohe, Zhumadian, Zhoukou and Xuchang, as well as the north of Zhengzhou, accounting for 43% of the total area of Henan Province. High risk area is mainly caused by high hazard of the flat terrain and abundant rainfall. The lower-risk area is mainly distributed in Sanmenxia, Jiyuan and Anyang, and the western part of Luoyang, that locate in high-elevation areas, and the vulnerability index of GDP is low in these areas. The rest of the areas are medium-risk areas. This research provides the background information for disaster prevention and mitigation management in Henan Province.