First, SAR images are often difficult to visually interpret, especially in dense urban areas. As illustrated in Fig. 1.1, it is hard to determine the location of streets, the boundaries of buildings or to identify individual buildings (e.g. to find the two towers of the Frauenkirche in the SARimage). This is related to the distortion effects pertinent to the SAR imaging concept. The layover effects lead to a mixture of backscatter from different objects at the same position in the SAR images; the shadow effects make many objects invisible; multiple scattering leads to bright lines, point signatures or even ghost scatterers (Auer et al. 2011) and causes high local contrasts in intensity. Man-made objects with different heights, shapes, materials or surface roughness appear in SAR images very differently, which also leads to unclear object boundaries. Nevertheless, exploiting these effects may bring us information which may be not contained in other kind ofdata (e.g. optical images or LiDARdata). For example, point signatures are strong hints of buildings (Soergel et al. 2006) and provide information about façade details such as windows or balconies (Auer et al. 2010a). Bright lines caused by double reflection signals indicate the boundaries of buildings (Wegner et al. 2010; Auer and Gernhardt 2014).
provided in literature. This dissertation contributes a pixel-based algorithm to detect increased backscattering in SAR images by analyzing the SAR pixel values according to simulated layers. To detect demolished buildings, simulated images are generated using LiDARdata. Two comparison operators (normalized mutual information and joint histogram slope) are used to compare image patches related to same buildings. An experiment using Munich data has shown that both of them provide an overall accuracy of more than 90%. A combinationof these two comparison operators using decision trees improves the result. The fourth objective is to detect changes between SAR images acquired with different incidence angles. For this purpose, three algorithms are presented in this dissertation. The first algorithm is a building-level algorithm based on layer fill. Image patches related to the same buildings in the two SAR images are extracted using simulation methods. For each extracted image patch pair, the change ratio based on the fill ratio of building layers is estimated. The change ratio values of all buildings are then classified into two classes using the EM-algorithm. This algorithm works well for buildings with different size and shape in complexurbanscenarios. Since the whole building is analyzed as one object, buildings with partly demolished walls may not be detected. Under the same idea, a wall-level changedetection algorithm was developed. Image patches related to the same walls in the two SAR images were extracted and converted to have the same geometry. These converted patch pairs are then compared using change ratios based on fill ratio or fill position. Lastly, the wall change results are fused to provide building change result. Compared to the building-level changedetection algorithm, this method is more time consuming, but yields better results for partly demolished buildings. A combinationof these two algorithms is therefore suggested, whereby the building-level method is used for all buildings and wall-level method additionally for selected large buildings. The third developed algorithm is a wall-level changedetection algorithm based on point-feature location. To this end, local maximum points in two SAR images corresponding to the same building façade are compared. This method provides promising result for the present data. It may work better for future datawith increased resolution to detect changes of detailed façade structures.
Nowadays spaceborne SARdata is easily available. Thanks to the high resolution of up to one meter (TerraSAR-X) it is suitable forurban applications, e.g. urban growth modeling as well as for damage mapping in conjunction with (natural) disasters. A main problem forSARimageinterpretation apart from the geometri- cal aspect is the high noise level caused by the combinationof deterministic (speckle effect) and random noise. The reduction of noise, e.g. by the multi-looking approach, often goes along with a loss of resolution. While structure preserving ﬁlters do not enhance ﬁne-structured areas, smoothening ﬁlters even blur the structures apparent in SARdata over urban areas. So reso- lution and structure preserving ﬁlter algorithms are still a topic of research. In this context alternative image representations like wavelets have been applied. While wavelets are used to separate point singularities (Cand`es and Donoho, 1999), second genera- tion wavelets, e.g. curvelets, are more suitable for the extraction of two dimensional features, as they are able to describe image discontinuities along a smooth line (an edge) with a minimum number of coefﬁcients (Cand`es and Donoho, 1999). The ele- mentary components are the so-called ridgelets – due to their appearance like a ridge – that can have different scales (equiv- alent to their length), directions and positions in the image. This enables a selection of two dimensional features to be suppressed (assumed noise) or to be emphasized (structure) by manipulating the corresponding coefﬁcient of each ridgelet. In the following a short overview to related work especially to the development of curvelets is given. Then, the curvelet representation is roughly explained and three applications are presented: image denoising, structure enhancement andchangedetection over the city center of Munich (imaged by TerraSAR-X in the high resolution spot- light mode and VV polarization). So this paper shows the poten- tial of the curvelet transform forSARimage analysis.
Various simulators have been developed to ease the interpretationofSAR images ofurban areas, e.g. by taking into account the electromagnetic and geometrical properties of buildings (Guida et al., 2008) or by utilizing ray tracing (Hammer and Schulz, 2011). GeoRaySAR is a simulator of the latter type, being de- veloped at DLR, which enables the identification of buildings in high resolution SARdata. To this end, prior knowledge about the scene geometry has to be extracted for the automated predic- tion of building extents. The knowledge can be acquired from either 3D GIS models (Auer and Donaubauer, 2015) or from DSMs (Tao et al., 2014). In (Tao et al., 2014) has prior knowl- edge been derived from DSMs based on LiDARdata (airborne sensor) that only contained man-made structures (i.e. vegetation had been pre-filtered). However, more realistic scenarios would expect DSMs on the basis of satellite data without pre-filtered vegetation (e.g. with support of cadastre information).
of C 0 would be lower than that of B 0 . On the right of figure 5, an example is given for the distribution ofSARimage signatures in case of a gable roof building. First, the signal response from the ground in front of the building, the building front wall and the roof is imposed within a layover area indicated by a box in bright gray pointing in range (range interval between dashed lines #1 and #2). Double reflections of signals, commonly referred to as double bounce, occur at the rectangular corner formed by the building front wall and the ground in front of the building (intersected by dashed line #2). The path of the double reflected radar signal is equal to the spatial distance between the SAR sensor and the corner tip marked by a bright circle situated on line #2. Due to the summary of double bounce responses within one resolution cell, a highlight occurs in range indicated by a bright box. Parts of the area behind the building are invisible to the SAR sensor and cause shadow. The shadow zone ends at the dashed line #3 when first reflections backscattered from the ground in the back of the building are detected. Changing the local angle of incidence of the radar signal or the aspect angle with respect to the building may completely change the geometrical appearance and radiometry of buildings in SARdata. Especially, ambiguities with regard to building parameters or misinterpretation may occur in the layover area where the distribution of scatterers in elevation is not provided. Case studies simulating basic shapes confirm the interpretation problem, e.g. using a box model (see figure 6a). Both the angle of incidence of the radar signal, which is 45 ◦ , and the aspect angle with respect to the box model are indicated by an arrow. The simulated reflectivity map, located in the azimuth-range plane, is displayed in figure 6b. Interpreting the map top-down in range, a zone of layover is followed by a bright double bounce line and a shadow zone of limited extent. Figure 6d shows simulation results for the same imaging geometry in case of a step model (see figure 6c) including two corners oriented in direction to the sensor. The only significant difference between the two reflectivity maps is found in the extent of the shadow area. Since shadow zones ofurban buildings are frequently imposed by the signal response from adjacent buildings, the extraction of object geometry is difficult. A closer look reveals differences in the intensity of double bounce. However, building extraction based on radiometrical information is challenging as well, as the backscattered power simultaneously depends on the dimension of objects and on object materials.
VHR SAR sensors such as TerraSAR-X (Pitz and Miller, 2010) or COSMO-SkyMed (Lom- bardo, 2004) provide SARdata having a spatial resolution of up to 1 m. In that kind ofdata, radar signals representing single objects are distributed over a high number of resolution cells. Hence, SARimage signatures corresponding to man-made objects become visible. For instance, in Bamler and Eineder (2008) salient SARimage features respresenting the Cheops pyramid in VHR spotlight TerraSAR-X data are introduced and interpreted. Figure 2 contains two SAR images of an urban area in Las Vegas, USA, comprising urban settlements, vegetated areas and the Las Vegas Convention Center. The images have been provided by the ERS satellite and TerraSAR-X in VHR spotlight mode. Captured in C-Band, the spatial resolution of the ERS data is 5 m x 25 m in azimuth and range, respectively. The signal to clutter ratio is much higher in the TerraSAR-X image ending up at a higher contrast in the high resolution data. Moreover, the number of signatures representing buildings in the TerraSAR-X data is much higher due to the increased resolution in azimuth and range. While distinguishing single ob- jects in the ERS data is almost impossible, buildings andurban structures are clearly visible on the TerraSAR-X image. Hence, data from spaceborne VHR SAR sensors become interesting for buidling extraction or changedetection methods using SAR amplitude data. In (Adam et al., 2008), the increase of the number of salient signatures in TerraSAR-X dataofurban areas is reported, also showing that the increase of the signal to clutter ratio provides better phase stability for dominant scatterers. Thus, monitoring of deformations of single objects based on phase information is enabled (Gernhardt et al., 2010; Zhu and Bamler, 2009).
Due to the all-weather data acquisition capabilities, high resolution space borne Synthetic Aperture Radar (SAR) plays an important role in remote sensing applications like changedetection. However, because of the complex geometric mapping of buildings in urban areas, SAR images are often hard to interpret. SARsimulationtechniques ease the visual interpretationofSAR images, while fully automatic interpretation is still a challenge. This paper presents a method for supporting the interpretationof high resolution SAR images with simulated radar images using a LiDAR digital surface model (DSM). Line features are extracted from the simulated and real SAR images and used for matching. A single building model is generated from the DSM and used for building recognition in the SARimage. An application for the concept is presented for the city centre of Munich where the comparison of the simulation to the TerraSAR-X data shows a good similarity. Based on the result ofsimulationand matching, special features (e.g. like double bounce lines, shadow areas etc.) can be automatically indicated in SARimage.
Junyi Tao was born in Wuhan, P.R.China, in 1983. He received the Dipl.-Ing.(Univ.) degree in Geodesy and Geoinformatics from the Universit¨at Stuttgart, Germany, in 2009, and the Dr.-Ing. degree in Geodesy and Geoinformatics from Technische Universit¨at M¨unchen, Munich, Germany, in 2015. From October 2009 to 2014, he was a scientific collaborator with the Remote Sensing Technology Institute (IMF), German Aerospace Center (DLR), Oberpfaffenhofen, Germany, working in close co- operation with the chair of Remote Sensing Tech- nology (LMF), Technische Universit¨at M¨unchen (TUM), Munich, Germany. In June 2012, he was a guest scientist at the Remote Sensing Laboratory, Department of Information Engineering and Computer Science, University of Trento, Trento, Italy. He pursued research topics in the field ofSARsimulation, SARimageinterpretation, multi-modal data fusion, andchangedetection. In particular, he focused on the combinationofLiDARandSARdatawithsimulationtechniquesfor object identification andchangedetection in urban areas. Currently, he is working in the R&D Department of a privately held company, where he is further developing his expertise in sensor data fusion.
In this paper, we discuss changedetection between a DSM as reference with an earlier acquisition dataand a high resolution SARimage (3D – 2D). Because of the all-weather and all-time data acquisition capability, SAR images are often the only available data in crisis situations, e.g. directly after an earthquake. For some cities, DSMs have been derived from LiDARdata or high resolution airborne optical imagedata before the disaster. Changedetection between these two data sets may enable a fast analysis of the damage caused by the disaster within the urban area and may lead to an improved rescue management in the disaster area.
The refined segmentations, as shown in Figure 5-6, are representing the buildings well. Some buildings are however still consisting out of many parts. This is due to large differences within the segments of a building, which can be seen in marked with a red circle in Figure 6-1. At this marked building, the corners of the building are very different from the middle part. While some times the multispectral channels can help to find similarities between the segments, at that building, all input channels are indicating an object change. For that, it is not possible to merge these together, as using a lower threshold would cause undersegmentation errors at other places. Concluding, oversegmentation cannot be completely eliminated from the segmentation.
In this paper, a method for supporting the visual interpretationofSAR images with simulated optical andSAR images using LiDAR DEM has been presented. Since the location and shape of the objects are similarly represented in the simulated images, acquiring the semantic on a SARimage is eased. The simulated optical image can be used for direct and quick identification of objects in the SARimage. The simulated SARimage has a similar signal reflectivity as the TerraSAR-X image, and it can also separately present single or multiple scattering in the SAR images, which is very useful for building recognition and reconstruction. Both the simulated optical andSAR images are automatically geocoded and enable a direct comparison with the SAR images. Future work will concentrate on learning the semantic relationship between objects in optical andSAR images in order to improve, for instance, methods forchangedetection.
Joshi et al.  and Yoo et al.  describe another approach for classification dependent compression and the achievable gain. Here, classification does not mean assignment of physical object classes, but assignment of different data classes, according to their statistic. However, to gain the high adaptability to the user requirement, class assignment to physical object classes is necessary. Advantageously, data belonging to one object class have similar statistic and thus necessary parameters for optimal quantization are easier to choose. Up to now, we split these classes also in small data blocks as in FLECS standard mode for two reasons: Similar design for all modes makes decompression easy, which will take place at the client; data statistics are too heterogeneous for a single quantizer. Optimization of the pre-classification with respect to data compression can lead to improved results.
Abstract: Mapping of landslides, quickly providing information about the extent of the affected area and type and grade of damage, is crucial to enable fast crisis response, i.e., to support rescue and humanitarian operations. Most synthetic aperture radar (SAR) data-based landslide detection approaches reported in the literature use changedetectiontechniques, requiring very high resolution (VHR) SAR imagery acquired shortly before the landslide event, which is commonly not available. Modern VHR SAR missions, e.g., Radarsat-2, TerraSAR-X, or COSMO-SkyMed, do not systematically cover the entire world, due to limitations in onboard disk space and downlink transmission rates. Here, we present a fast and transferable procedure for mapping of landslides, based on changedetection between pre-event optical imagery and the polarimetric entropy derived from post-event VHR polarimetric SARdata. Pre-event information is derived from high resolution optical imagery of Landsat-8 or Sentinel-2, which are freely available and systematically acquired over the entire Earth’s landmass. The landslide mapping is refined by slope information from a digital elevation model generated from bi-static TanDEM-X imagery. The methodology was successfully applied to two landslide events of different characteristics: A rotational slide near Charleston, West Virginia, USA and a mining waste earthflow near Bolshaya Talda, Russia.
ACCIDENT DATA GIDAS database
The major objective in assessing accident avoidance potentials is a precise assessment as to the positive potential of a driver assistance technology on accident scenarios in urban areas. This assessment requires accident data containing relevant accident parameters in order to determine the influence of functional limits (weather, road layout). Official police recorded accident statistics do not cover aspects such as collision speed or pre-crash trajectory, hence the accident data are limited and will not allow statistical traffic accident overviews. GIDAS (German In-Depth Accident Study) was established in 1999 as a cooperation project between the Federal Highway Research Institute Germany (BASt) and the research association on automotive engineering of German Car Industry (FAT). In- depth data in GIDAS combines data collection at the accident scene and the time of the accident applying retrospective methods like measuring, collection of medical dataand accident reconstruction. The general concept of GIDAS is to compile a random, un-biased and representative sample of 2,000 accidents involving injury per year to cover all parameters of German road accident scenarios. No pre- selection of severe cases is performed, but all accidents involving injuries as defined by the police are considered. The accident reconstruction is conducted for each GIDAS case, a unique feature of GIDAS, since information for accidents of all severity classes  is available. The project UR:BAN focuses on the pre-crash matrices describing trajectories and kinematics of involved vehicles and pedestrians and therefore allow using reconstruction information of real accidents for the virtual simulationof driver assistance technologies during the development of these features.
Differential Interferometric SAR (DInSAR) and multitemporal DInSAR have proven to be very effective in mapping surface displacement . The TerraSAR-X mission provides meter- resolution imagery data from space and has provided a significant improvement in the monitoring capabilities of interferometric techniques . Due to reduced wavelength (ca. 3.1 cm), the data are more sensitive to small displacements, even those caused by thermal dilation of objects. However, most of the stacking algorithms, such as Persistent Scatterer Interferometry (PSI), estimate the deformation time series assuming a linear subsidence . This is especially not true in the case of man-made structures, such as steel core bridges and specific buildings, which may be very sensitive to thermal dilation effects correlated to the temperature and may undergo large non-linear seasonal deformation. Thus, assumption of a linear model may often lead to ambiguous velocity estimates and decreased performance. Thus, a DInSAR stacking system dealing with non-linear motion has to be adapted accordingly. The aim of this paper is thermal dilation monitoring ofcomplexurban structures at high resolution by extending standard multitemporal DInSAR techniques based on
el based algorithm is performed to detect positive changes of large extent in shadow and ground areas. Based on experimental results, an intensity threshold is determined by a statistical analysis of the whole SARimage, what is done for each image layer. All pixels in the corresponding layer in the SARimage are compared to this threshold value and are then classified to ‘change’ and ‘no change’ according to Table 2. The detailed steps are described in . Exploiting the layover and double bounce layers, we tried the same method based on the assumption that low intensity pixels indicate negative changes. How- ever, the reflection of the radar signals depends on many different physical parameters (e.g. material, roughness), which may be only roughly considered in the simulation step and are not provided by the LI- DAR DSM. Moreover, due to the lack of geometrical information in the 2.5D DEM, no facade signal can be simulated in order to predict the appearance of salient signatures. To conclude, the amount of a-priori knowledge provided by the DEM does not suffice for a generalized pixel-based analysis for negative changes, as there are many “false alarms” in the changedetection result (also in case of a significant reduction of spatial resolution). Hence, object infor- mation has to be included in addition to the 2.5D DEM, e.g. facade grammar characterizing buildings in the local scene of interest.
One example parameter trade map is shown in Figure 8 and is explained in the following starting from lower left. The limiting factor for the swath is a timing issue constraint by the available echo window length. Multi- SCORE technique can be used to increase the swath width; here multiple digital beams are generated, each tracking the ground echo of –otherwise– ambiguous sig- nals. However, the total wide swath will contain gaps. To overcome this, the pulse-to-pulse interval is varied over time , a technique known as staggered PRF. This
The goal of this paper is to introduce and share a dataset derived from the SARptical framework. The dataset consists of over 10,000 pairs of corresponding SARand optical image patches extracted from TerraSAR-X high resolution
6.1 Training data
all tiles according to the tile map service (TMS) specification  are processed. This includes the download of the corresponding orthophoto from Bing Maps and the creation of at least one binary image. The binary image contains all ground-truth instances for the current class – such as building or highway – that was specified in the configuration. This step creates one satellite imageand one or more binary images per tile. If, for the current tile, no ground truths according to the configuration are found, no images for this tile are created. Figure 6.1 shows the output of a single tile, generated by Airtiler. An important point is the file names. These names are generated from the content of the file, which makes the loading andinterpretationof the data set rather simple. Image names include the file extension .tiff whereas the masks have the extension .tif.
The capabilities of classical single-channel SAR systems regarding wide-swath imaging with high azimuth resolution are inherently limited by contradicting pulse repetition frequency (PRF) requirements. Wide swathes require large echo windows and therefore a small PRF. However, the wide Doppler spectrum of high-resolution images demands a high PRF in order to adequately sample the signal respecting the Nyquist criterion. Violations of the sampling criterion lead to rising azimuth ambiguities in the imageand reduce the interpretability of the data products. Using multiple phase centers offers the ability to overcome the sampling constraints and acquire high-resolution wide-swath images , . Spaceborne experiments with two channels of TerraSAR-X  operated in the “dual receive antenna” (DRA)  mode and the TerraSAR-X and TanDEM-X formation flying system  have been reported in the literature. Also a four channel experiment with TerraSAR-X and TanDEM-X jointly operating in DRA mode has been conducted . These experimental datasets were acquired during a dedicated science phase  which was conducted after the completion of the primary mission goal of TanDEM-X, the acquisition of a global digital elevation model (DEM) with outstanding accuracy . During the science phase the geometry of the