No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Refinement of Digital Elevation Models from Shadowing Cues
Abstract
In this paper we derive formal constraints relating terrain elevation and observed cast shadows. We show how an optimisation framework can be used to refine surface estimates using shadowing constraints from one or more images. The method is particularly applicable to the digital elevation models produced by the Shuttle Radar Topography Mission (SRTM), which have an abundance of voids in mountainous areas where elevation data is missing. Cast shadow maps are detected automatically from multi-spectral satellite imagery using a simple heuristic which is reliable over varying types of surface cover. We show that the combination of our shadow segmentation and terrain correction methods can restore the structure of mountain ridges in interpolated SRTM voids using five satellite images, decreasing the RMS error by over 25%.
... However, the limitation of this method is that it cannot be extended to the DEM data with different resolutions and auxiliary DEM data are not always available. Hogan et al. [8] improved interpolation results according to geometric constraints provided by shadows. However, shadow geometric constraints obtained from shadows provide relatively sparse information and the method is accompanied by a nonconvex optimization problem which is difficult to compute and may fall into a local optimum. ...
... We begin by explaining how we automatically detect shadow areas in terrain imagery. We employ the multi-band thresholding technique [8] to perform shadow segmentation. Specifically, we use three bands (near infrared, mid-infrared and thermal infrared bands) of multispectral satellite images from Landsat-5. ...
... Computing this for every pixel leads to a binary shadow map of the same size as the image, as shown in Fig. 3. The segmentation error is less than 10% which is empirically validated to be acceptable for shadow characterization [8]. ...
We explore the use of convolutional neural networks (CNNs) for filling voids in digital elevation models (DEM). We propose a baseline approach using a fully convolutional network to predict complete from incomplete DEMs, which is trained in a supervised fashion. We then extend this to a shadow-constrained CNN (SCCNN) by introducing additional loss functions that encourage the restored DEM to adhere to geometric constraints implied by cast shadows. At the training time, we use automatically extracted cast shadow maps and known sun directions to compute the shadow-based supervisory signal in addition to the direct DEM supervision. At the test time, our network directly predicts restored DEMs from an incomplete DEM. One key advantage of our SCCNN model is that it is characterized by both CNN data inference and geometric shadow cues. It thus avoids data restoration that may violate shadowing conditions. Both our baseline CNN and SCCNN outperform the inverse distance weighting (IDW)-based interpolation method, with the shadow supervision enabling SCCNN to obtain the best performance.
... However, the accuracy of SRTM data tends to be influenced by various environmental situations. For example, the areas where the reflectivity of radar signal are weak cannot be imaged [1,2], resulting in voids where elevation data is unavailable. Additionally, blurry and incomplete data normally arise in depicting high altitude mountains where the terrain is complex. ...
... In the light of this observation, they proposed a method that uses different algorithms to restore voids of different sizes and types. Hogan et al. [2] proposed a method by using shadow cues to recover SRTM void data. They analyzed the relationships between shadows and terrain elevations, and used shadows to constrain the void-filling process. ...
... Though the interpolation strategies are capable of effectively filling small void areas, their performance tremendously decreases in the case of large voids. One reason for this effectiveness is that the interpolation strategies tend to employ either exterior information such as shadows [2] and Digital Elevation Maps or interior neighboring elements for guiding the void filling, but rarely exploit the correlations between void data and existing data within one image. However, the terrains that are not imaged and exhibit voids in one DEM do have close accompanying relationships with those existing data in the DEM. ...
... They utilized a DSM from high resolution optical stereo imagery and the SRTM DEM. Hogan et al. (2010) proposed a method to fill the voids in SRTM using shadowing cues detected from multispectral satellite image. They derived a geometric shadow constraint from one or more images and in combination of terrain correction methods for refining digital elevation model in mountainous area. ...
... For shadow detection from multispectral satellite images, many algorithms have been proposed. Generally as the shadow increases, red and NIR reflectance decreases but the normalised difference between the two increases (Hogan et al., 2010). The formula proposed in (Shahi et al, 2014) is used here to automatically detect shadows. ...
Digital Surface Models (DSM) derived from stereo-pair satellite images are the main sources for many Geo-Informatics applications like 3D change detection, object classification and recognition. However since occlusion especially in urban scenes result in some deficiencies in the stereo matching phase, these DSMs contain some voids. In order to fill the voids a range of algorithms have been proposed, mainly including interpolation alone or along with auxiliary DSM. In this paper an algorithm for void filling in DSM from stereo satellite images has been developed. Unlike common previous approaches we didn’t use any external DSM to fill the voids. Our proposed algorithm uses only the original images and the unfilled DSM itself. First a neighborhood around every void in the unfilled DSM and its corresponding area in multispectral image is defined. Then it is analysed to extract both spectral and geometric texture and accordingly to assign labels to each cell in the voids. This step contains three phases comprising shadow detection, height thresholding and image segmentation. Thus every cell in void has a label and is filled by the median value of its co-labelled neighbors. The results for datasets from WorldView-2 and IKONOS are shown and discussed.
... They utilized a DSM from high resolution optical stereo imagery and the SRTM DEM. Hogan et al. (2010) proposed a method to fill the voids in SRTM using shadowing cues detected from multispectral satellite image. They derived a geometric shadow constraint from one or more images and in combination of terrain correction methods for refining digital elevation model in mountainous area. ...
... For shadow detection from multispectral satellite images, many algorithms have been proposed. Generally as the shadow increases, red and NIR reflectance decreases but the normalised difference between the two increases (Hogan et al., 2010). The formula proposed in (Shahi et al, 2014) is used here to automatically detect shadows. ...
Digital Surface Models (DSM) derived from stereo-pair satellite images are the main sources for many Geo-Informatics applications like 3D change detection, object classification and recognition. However since occlusion especially in urban scenes result in some deficiencies in the stereo matching phase, these DSMs contain some voids. In order to fill the voids a range of algorithms have been proposed, mainly including interpolation alone or along with auxiliary DSM. In this paper an algorithm for void filling in DSM from stereo satellite images has been developed. Unlike common previous approaches we didn’t use any external DSM to fill the voids. Our proposed algorithm uses only the original images and the unfilled DSM itself. First a neighborhood around every void in the unfilled DSM and its corresponding area in multispectral image is defined. Then it is analysed to extract both spectral and geometric texture and accordingly to assign labels to each cell in the voids. This step contains three phases comprising shadow detection, height thresholding and image segmentation. Thus every cell in void has a label and is filled by the median value of its co-labelled neighbors. The results for datasets from WorldView-2 and IKONOS are shown and discussed.
... They utilized a DSM from high resolution optical stereo imagery and the SRTM DEM. Hogan et al. (2010) proposed a method to fill the voids in SRTM using shadowing cues detected from multispectral satellite image. They derived a geometric shadow constraint from one or more images and in combination of terrain correction methods for refining digital elevation model in mountainous area. ...
... For shadow detection from multispectral satellite images, many algorithms have been proposed. Generally as the shadow increases, red and NIR reflectance decreases but the normalised difference between the two increases (Hogan et al., 2010). The formula proposed in (Shahi et al, 2014) is used here to automatically detect shadows. ...
Digital Surface Models (DSM) derived from stereo-pair satellite images are the main sources for many Geo-Informatics applications like 3D change detection, object classification and recognition. However since occlusion especially in urban scenes result in some deficiencies in the stereo matching phase, these DSMs contain some voids. In order to fill the voids a range of algorithms have been proposed, mainly including interpolation alone or along with auxiliary DSM. In this paper an algorithm for void filling in DSM from stereo satellite images has been developed. Unlike common previous approaches we didn’t use any external DSM to fill the voids. Our proposed algorithm uses only the original images and the unfilled DSM itself. First a neighborhood around every void in the unfilled DSM and its corresponding area in multispectral image is defined. Then it is analysed to extract both spectral and geometric texture and accordingly to assign labels to each cell in the voids. This step contains three phases comprising shadow detection, height thresholding and image segmentation. Thus every cell in void has a label and is filled by the median value of its co-labelled neighbors. The results for datasets from WorldView-2 and IKONOS are shown and discussed.
... It is of great interest to properly consider external auxiliary data to improve the accuracy of DEM void filling. The selection of auxiliary data is a key factor, such as extracting valley lines from Landsat sensor imagery [19], night time ASTER thermal imagery data [20], and shadow maps from multispectral images [21,22].In fact, these auxiliary data are not homogeneous with those of DEMs, and are not as simple and direct as DEM. External DEMs in the same area are the most commonly used auxiliary data [13,[23][24][25]; the classic one is the fill and feather (F&F) method [15]. ...
Accurate and complete digital elevation models (DEMs) play an important fundamental role in geospatial analysis, supporting various engineering applications, human activities, and scientific research. Interferometric synthetic aperture radar (InSAR) plays an increasingly important role in DEM generation. Nonetheless, owing to its inherent characteristics, gaps often appear in regions marked by significant topographical fluctuations, necessitating an extra void-filling process. Traditional void-filling methods have operated directly on preexisting data, succeeding in relatively flat terrain. When facing mountainous regions, there will always be gross errors in elevation values. Regrettably, conventional methods have often disregarded this vital consideration. To this end, this research proposes a DEM void-filling method based on incorporating elevation outlier detection. It accounts for the detection and removal of elevation outliers, thereby mitigating the shortcomings of existing methods and ensuring robust DEM restoration in mountainous terrains. Experiments were conducted to validate the method applicability using TanDEM-X data from Sichuan, China, Hebei, China, and Oregon, America. The results underscore the superiority of the proposed method. Three traditional methods are selected for comparison. The proposed method has different degrees of improvement in filling accuracy, depending on the void status of the local terrain. Compared with the delta surface fill (DSF) method, the root mean squared error (RMSE) of the filling results has improved by 7.87% to 51.87%. The qualitative and quantitative experiments demonstrate that the proposed method is promising for large-scale DEM void-filling tasks.
The digital elevation model (DEM) acquired through photogrammetry or LiDAR usually exposes voids due to phenomena such as instrumentation artifact, ground occlusion, etc. For this reason, this paper proposes a multiattention generative adversarial network model to fill the voids. In this model, a multiscale feature fusion generation network is proposed to initially fill the voids, and then a multiattention filling network is proposed to recover the detailed features of the terrain surrounding the void area, and the channel-spatial cropping attention mechanism module is proposed as an enhancement of the network. Spectral normalization is added to each convolution layer in the discriminator network. Finally, the training of the model by a combined loss function, including reconstruction loss and adversarial loss, is optimized. Three groups of experiments with four different types of terrains, hillsides, valleys, ridges and hills, are conducted for validation of the proposed model. The experimental results show that (1) the structural similarity surrounding terrestrial voids in the three types of terrains (i.e., hillside, valley, and ridge) can reach 80–90%, which implies that the DEM accuracy can be improved by at least 10% relative to the traditional interpolation methods (i.e., Kriging, IDW, and Spline), and can reach 57.4%, while other deep learning models (i.e., CE, GL and CR) only reach 43.2%, 17.1% and 11.4% in the hilly areas, respectively. Therefore, it can be concluded that the structural similarity surrounding the terrestrial voids filled using the model proposed in this paper can reach 60–90% upon the types of terrain, such as hillside, valley, ridge, and hill.
The original data produced by the Shuttle Radar Topography Mission (SRTM) tend to have an abundance of voids in mountainous areas where the elevation measurements are missing. In this paper, deep learning models are investigated for restoring SRTM data. To this end, we explore generative adversarial nets, which represent one state-of-the-art family of deep learning models. A conditional generative adversarial network (CGAN) is introduced as the baseline method for filling voids in incomplete SRTM data. The problem regarding shadow violation that possibly arises from the CGAN restored data is investigated. To address this deficiency, shadow geometric constraints based on shadow maps of satellite images are devised. In addition, a shadow constrained conditional generative adversarial network (SCGAN), which incorporates the shadow geometric constraints into the CGAN, is developed. Training the SCGAN model requires both the remote sensing observations (i.e., the original incomplete SRTM data and satellite images) and the ground truth data (i.e., the complete SRTM data, which are manually refined from the incomplete SRTM data with the reference of in-situ measurements). The integration of the multi-source training data enables the SCGAN model to be characterized by comprehensive information including both mountain shape variation and mountain shadow geometry. Experimental results validate the superiority of the SCGAN over the comparison methods, i.e., the interpolation, the convolutional neural network (CNN) and the baseline CGAN, in SRTM data restoration.
The Shuttle Radar Topography Mission (SRTM) delivered a free digital elevation model (DEM) with a spatial resolution of 3'' at near global coverage. However, there are many data voids in the SRTM data that need to be filled before their application. In this Letter, a novel void-filling method, which uses Landsat sensor imagery to provide detailed information on terrains for the filling procedure, is proposed. Valleys are extracted from Landsat sensor imagery first, then elevation values of SRTM voids along valleys are interpolated, and finally the remaining voids are patched with the global coarse resolution DEM. The test results show that the elevation values filled with the proposed method perform better than those filled with the traditional method and local geomorphological features can be maintained with the proposed method.
The Digital Elevation Model that has been derived from the February 2000 Shuttle Radar Topography Mission (SRTM) has been one of the most important publicly available new spatial data sets in recent years. However, the 'finished' grade version of the data (also referred to as Version 2) still contains data voids (some 836,000 km 2)—and other anomalies—that prevent immediate use in many applications. These voids can be filled using a range of interpolation algorithms in conjunction with other sources of elevation data, but there is little guidance on the most appropriate void-filling method. This paper describes: (i) a method to fill voids using a variety of interpolators, (ii) a method to determine the most appropriate void-filling algorithms using a classification of the voids based on their size and a typology of their surrounding terrain; and (iii) the classification of the most appropriate algorithm for each of the 3,339,913 voids in the SRTM data. Based on a sample of 1304 artificial but realistic voids across six terrain types and eight void size classes, we found that the choice of void-filling algorithm is dependent on both the size and terrain type of the void. Contrary to some previous findings, the best methods can be generalised as: kriging or inverse distance weighting interpolation for small and medium size voids in relatively flat low-lying areas; spline interpolation for small and medium-sized voids in high-altitude and dissected terrain; triangular irregular network or inverse distance weighting interpolation for large voids in very flat areas, and an advanced spline method (ANUDEM) for large voids in other terrains.
The Digital Elevation Model that has been derived from the February 2000 Shuttle Radar Topography Mission (SRTM) has been one of the most important publicly available new spatial data sets in recent years. However, the ‘finished’ grade version of the data (also referred to as Version 2) still contains data voids (some 836,000 km)—and other anomalies—that prevent immediate use in many applications. These voids can be filled using a range of interpolation algorithms in conjunction with other sources of elevation data, but there is little guidance on the most appropriate void‐filling method. This paper describes: (i) a method to fill voids using a variety of interpolators, (ii) a method to determine the most appropriate void‐filling algorithms using a classification of the voids based on their size and a typology of their surrounding terrain; and (iii) the classification of the most appropriate algorithm for each of the 3,339,913 voids in the SRTM data. Based on a sample of 1304 artificial but realistic voids across six terrain types and eight void size classes, we found that the choice of void‐filling algorithm is dependent on both the size and terrain type of the void. Contrary to some previous findings, the best methods can be generalised as: kriging or inverse distance weighting interpolation for small and medium size voids in relatively flat low‐lying areas; spline interpolation for small and medium‐sized voids in high‐altitude and dissected terrain; triangular irregular network or inverse distance weighting interpolation for large voids in very flat areas, and an advanced spline method (ANUDEM) for large voids in other terrains.
The Shuttle Radar Topography Mission produced the most complete, highest-resolution digital elevation model of the Earth. The project was a joint endeavor of NASA, the National Geospatial-Intelligence Agency, and the German and Italian Space Agencies and flew in February 2000. It used dual radar antennas to acquire interferometric radar data, processed to digital topographic data at 1 arc sec resolution. Details of the development, flight operations, data processing, and products are provided for users of this revolutionary data set.
The goal of this study was to characterize and quantify the occurrence of data voids in data from the Shuttle Radar Topography Mission (SRTM) for the conterminous United States. For this purpose, SRTM data and corresponding data from the national elevation data were downloaded in 21 samples spatially organized to cover the main topography of the U.S. Void locations in SRTM data were compared to terrain attributes and subsequently the area of individual data voids to the same attributes. It was found that data voids amounted to 0.3% of the total dataset. Data voids were found in all topographic settings but more often in slopes steeper than approximately 20° that face south (170°), and also in flat areas such as lakes and rivers. It was also found that more than 50% of all data voids were composed of connected pixels in groups less than six pixels. The largest data voids could be attributed to water bodies, while the rest could be explained by terrain-radar interaction characteristics.
This article presents a new technique for filling voids in SRTM DEM data. Tests prove the increased effectiveness of DSF in filling SRTM voids compared to the F&F technique, especially at the problematic void interfaces. DSF gives better results by both visual and quantitative measures. Because of these performance improvements, DSF is now in use by NGA and its contractors in SRTM void filling. In addition to operations on SRTM datasets, DSF has application to other DEM-level void filling endeavors. Numerous combinations of elevation data can be used in the parent and fill surface roles.
The primary goal of the session Integration of Digital Elevation Data into Hydrologic Models, sponsored by the Surface Water Committee of the AGU Hydrology Section at the AGU 1989 Fall Meeting in San Francisco, was to provide an overview of the various types of Digital Elevation Model (DEM) data and methods used to incorporate these data into rainfall-runoff models.
We design an algorithm based on illuminant invariance theory to find shadow regions in a colour image. Shad- ows are caused by a local change in both the colour and the intensity of illumination. Using both chromaticity and intensity cues, an illuminant discontinuity measure is de- rived by which shadow edges can be locally identified. We model the problem of finding shadows by a Markov Random Field using our new measure. A graph-cut opti- mization method is then applied to the MRF to find the globally optimal segmentation of shadows in an image. In previous work, a 2-d chromaticity colour invariant im- age was recovered from a greyscale 1-d invariant image by adding back light so as to match the chromaticity of bright pixels. Here, since we segment shadows, we can take a completely different approach and leave nonshadow pix- els unchanged, while adding light to shadow pixels so as to match neighbouring nonshadow pixels. The results are much more convincing shadow-free images, and shadow- segmentation is excellent.
We present an algorithm for performing Lambertian pho- tometric stereo in the presence of shadows. The algorithm has three novel features. First, a fast graph cuts based method is used to estimate per pixel light source visibility. Second, it allows images to be acquired with multiple illu- minants, and there can be fewer images than light sources. This leads to better surface coverage and improves the re- construction accuracy by enhancing the signal to noise ratio and the condition number of the light source matrix. The ability to use fewer images than light sources means that the imaging effort grows sublinearly with the number of light sources. Finally, the recovered shadow maps are combined with shading information to perform constrained surface normal integration. This reduces the low frequency bias inherent to the normal integration process and ensures that the recovered surface is consistent with the shadowing con- figuration The algorithm works with as few as four light sources and four images. We report results for light source visibil- ity detection and high quality surface reconstructions for synthetic and real datasets.
In this paper we discuss new results on the Shape From Darkness
problem: using the motion of cast shadows to recover scene structure.
Our approach is based on collecting a set of images from a fixed
viewpoint as a known light source mover; “across the sky”.
Previously published solutions to this problem have performed the
reconstruction only for cross sections of the scene. In this paper, we
present a reconstruction algorithm and discuss the reconstruction of an
entire 3-D scene under various light source trajectories. We also
consider the constraints on reconstruction. We conclude with
experimental results that illustrate the convergence properties of the
solution process and its robustness properties
The behavior of canopy reflectance and spectral bandtransformations, normalized difference (ND), and greenness (GR) dueto the amount and location of scene shadows was examined. Live balsamfir (Abies balsamea L. (Mill.)) trees were arranged in an equidistantpattern on a platform. Spectral reflectances were acquired forthree canopies (44-, 70-,and 90-pereent cover) and three backgroundtypes (grass-sod, black-,and white-painted boards) for a range of solarzenith angles. Rotating the platform produced variations in the locationof shadows cast by trees in the scene.Red and near-IR reflectance factors decreased as the amount of sceneshadows increased while ND increased with increasing scene shadows.However, when large anmounts of green vegetation were present, NDsaturated (i.e., approached an asymptote) and the observed increaseswere reduced. GR was least affected by changing shadow amounts.Spectral response variations due to shadows decreased as canopy coverincreased.A simple model was used to examine the effects of shadows in "pure"pixels of backgrounds representing vegetation, bare soil, and snow.Red and near-IR reflectance decreased and ND increased as the proprotionof shadows increased. These results suggest that green vegetationphytomass may be overestimated by ND when shadows cast byvegetation are present in a scene.