Can inpainting improve digital terrain analysis? Comparing techniques for void filling, surface reconstruction and geomorphometric analyses

Crema, Stefano; Llena, Manel; Calsamiglia, Aleix; Estrany, Joan; Marchi, Lorenzo; Vericat, Damià; Cavalli, Marco

doi:10.1002/esp.4739

Cited by 19 publications

(12 citation statements)

References 57 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…On the other hand, [39]- [41] assume the availability of land cover classes, i.e. from ground filtering, and compare various image inpainting methods, among which are the GANbased [42], in completing the DTMs.…”

Section: A Dtm Extractionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Digital Terrain Models From Point Clouds: ALS2DTM Dataset and Rasterization-Based GAN

Guiotte

Pham

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Despite the popularity of deep neural networks in various domains, the extraction of digital terrain models (DTMs) from airborne laser scanning (ALS) point clouds is still challenging. This might be due to the lack of dedicated large-scale annotated dataset and the data-structure discrepancy between point clouds and DTMs. To promote data-driven DTM extraction, this paper collects from open sources a large-scale dataset of ALS point clouds and corresponding DTMs with various urban, forested, and mountainous scenes. A baseline method is proposed as the first attempt to train a Deep neural network to extract digital Terrain models directly from ALS point clouds via Rasterization techniques, coined DeepTerRa. Extensive studies with well-established methods are performed to benchmark the dataset and analyze the challenges in learning to extract DTM from point clouds. The experimental results show the interest of the agnostic data-driven approach, with sub-metric error level compared to methods designed for DTM extraction. The data and source code is provided at https://lhoangan.github.io/deepterra/ for reproducibility and further similar research.

show abstract

Section: A Dtm Extractionmentioning

confidence: 99%

“…To maintain consistency with the previous setups, we concatenate semantic maps to the rasters as extra channels and leave the exhaustive study of efficient usage of semantics for future work. This is a less direct use of semantic information compared to the more popular way in which objects' regions are removed and inpainted [39]- [41].…”

Section: E Semanticsmentioning

confidence: 99%

Learning Digital Terrain Models From Point Clouds: ALS2DTM Dataset and Rasterization-Based GAN

Guiotte

Pham

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Terrestrial SfM can also be a good solution when flight permissions are not granted by local legislation at certain sites or when weather conditions may not be suitable for drone flights (e.g., high wind speeds [12]). However, the ground elevation under vegetation is usually not represented in SfM, limiting the quality of the resulting point cloud [55]. Another constraint for SfM surveys, both in terms of time and cost [56], especially in steep-slope and rugged environments, is the ground control point (GCP) requirement.…”

Section: Methodsmentioning

confidence: 99%

“…The statistics reported in Table 5 show how the interpolation process slightly increased the errors in DTM data compared to the errors at the point cloud level. It is well known that gridding processes may induce a loss of resolution and increased errors with respect to the original data [55,98,99,124,125], especially in complex topography [77]. However, the quality of the DTMs obtained was certainly high (centimeter-level), considering the values in Table 5, which demonstrated how the data fusion process proved effective in generating HRT data.…”

Section: Dtm Error Assesmentmentioning

confidence: 96%

Multiplatform-SfM and TLS Data Fusion for Monitoring Agricultural Terraces in Complex Topographic and Landcover Conditions

et al. 2020

View full text Add to dashboard Cite

Agricultural terraced landscapes, which are important historical heritage sites (e.g., UNESCO or Globally Important Agricultural Heritage Systems (GIAHS) sites) are under threat from increased soil degradation due to climate change and land abandonment. Remote sensing can assist in the assessment and monitoring of such cultural ecosystem services. However, due to the limitations imposed by rugged topography and the occurrence of vegetation, the application of a single high-resolution topography (HRT) technique is challenging in these particular agricultural environments. Therefore, data fusion of HRT techniques (terrestrial laser scanning (TLS) and aerial/terrestrial structure from motion (SfM)) was tested for the first time in this context (terraces), to the best of our knowledge, to overcome specific detection problems such as the complex topographic and landcover conditions of the terrace systems. SfM–TLS data fusion methodology was trialed in order to produce very high-resolution digital terrain models (DTMs) of two agricultural terrace areas, both characterized by the presence of vegetation that covers parts of the subvertical surfaces, complex morphology, and inaccessible areas. In the unreachable areas, it was necessary to find effective solutions to carry out HRT surveys; therefore, we tested the direct georeferencing (DG) method, exploiting onboard multifrequency GNSS receivers for unmanned aerial vehicles (UAVs) and postprocessing kinematic (PPK) data. The results showed that the fusion of data based on different methods and acquisition platforms is required to obtain accurate DTMs that reflect the real surface roughness of terrace systems without gaps in data. Moreover, in inaccessible or hazardous terrains, a combination of direct and indirect georeferencing was a useful solution to reduce the substantial inconvenience and cost of ground control point (GCP) placement. We show that in order to obtain a precise data fusion in these complex conditions, it is essential to utilize a complete and specific workflow. This workflow must incorporate all data merging issues and landcover condition problems, encompassing the survey planning step, the coregistration process, and the error analysis of the outputs. The high-resolution DTMs realized can provide a starting point for land degradation process assessment of these agriculture environments and supplies useful information to stakeholders for better management and protection of such important heritage landscapes.

show abstract

“…The capability of a DEM to reproduce terrain morphology cannot be fully captured by the statistical analysis of pixel-by-pixel differences respect to a reference DEM, adopting conventional metrics such as root mean square error (RMSE), mean/median error and mean/median absolute error (e.g., Florinsky et al, 2019;Crema et al 2020;Polidori & El Hage, 2020). Differently, the spatial variability structure (Isaaks & Srivastava, 1989) of DEMs should be considered in the evaluation of their quality, for example including the capability to reproduce ne-scale morphology and different aspects of surface roughness (e.g., Trevisani, Cavalli, & Marchi, 2012).…”

Section: Introductionmentioning

confidence: 99%

Global digital elevation models for terrain morphology analysis in mountain environments: insights on Copernicus GLO-30 and ALOS AW3D30 for a large Alpine area

2023

View full text Add to dashboard Cite

This study focuses on the quality evaluation of two of the best 1 arc-second public global digital elevation models (DEMs), Copernicus GLO-30 DEM and ALOS AW3D30 DSM, from the perspective of their capability to represent the terrain morphology of a complex alpine landscape, located in the alpine Trentino Province, in the Italian Alps. The analysis is performed on an area of 6210 km 2 , considering a reference DEM derived from a high resolution and accurate airborne Lidar DEM. The quality assessment includes, in addition to a conventional analysis of error statistics on a pixels-by-pixel basis, an ad-hoc analysis on the capability to represent the ne-scale morphology and local roughness. The quality analysis is performed considering the in uence of local morphology and of the different land covers. The ndings show that the two global DEMs have comparable overall quality, but the relative performances change according to local landscape characteristics. Copernicus DEM performance is on average better than ALOS in correspondence of urbanized areas as well in areas without vegetation cover, with gentle slopes and relatively low short-range roughness. ALOS DEM performance is slightly better thanCopernicus in rougher terrain and steeper slopes. In general, both DEMs have poor performances in steep slopes, with a limited capability to describe correctly local morphology. The adoption of these global DEMs for terrain analysis and modelling of earth surface processes should be performed carefully, taking into account the impact of different land covers and of local morphology, including surface roughness.

show abstract

Can inpainting improve digital terrain analysis? Comparing techniques for void filling, surface reconstruction and geomorphometric analyses

Cited by 19 publications

References 57 publications

Learning Digital Terrain Models From Point Clouds: ALS2DTM Dataset and Rasterization-Based GAN

Learning Digital Terrain Models From Point Clouds: ALS2DTM Dataset and Rasterization-Based GAN

Multiplatform-SfM and TLS Data Fusion for Monitoring Agricultural Terraces in Complex Topographic and Landcover Conditions

Global digital elevation models for terrain morphology analysis in mountain environments: insights on Copernicus GLO-30 and ALOS AW3D30 for a large Alpine area

Contact Info

Product

Resources

About