IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning
<p>Aerial images from the first dataset (left), the corresponding nDSMs in heat map format and their color bars, indicating the color-coding of the nDSM in meters (right). The aerial images on the left of each pair have a size of 4000 × 4000, while the size of the nDSMs is 1000 × 1000. nDSMs are presented at the same size as the aerial images for demonstration reasons. The Figure is best seen in color.</p> "> Figure 2
<p>The DSM (<b>left</b>) and the DTM (<b>right</b>) corresponding to the bottom aerial image of <a href="#remotesensing-13-02417-f001" class="html-fig">Figure 1</a>. The color bar for each heat map indicates the color-coding of the DEMs in meters above sea level. Both heat maps have several undetermined or irrational (extremely high or low) values shown in black color. Notably, some of these unexpected values in the DSM map (<b>left</b>) correspond to a river, which illustrates a well-known problem of LiDAR measurements near highly reflective and refractive surfaces with multiple light paths. Such erroneous values raise significant problems regarding the training of the model. Thus, they are detected during data preprocessing and excluded from the training data (see <a href="#sec2dot4-remotesensing-13-02417" class="html-sec">Section 2.4</a>). They are also excluded from the validation and test data to avoid inaccurate performance evaluations. Overall, these values roughly comprise 10% of the dataset but lead to a larger amount of discarded data, since any candidate patch containing even a pixel of undetermined or irrational value is excluded from the training pipeline. This figure is best seen in color.</p> "> Figure 3
<p>Aerial images from the IEEE GRSS Data Fusion Contest (second dataset), the corresponding nDSMs and the color bars of the heat maps indicating the color-coding in meters. The RGB images on the left of each pair have a size of 5000 × 5000 pixels, while the size of the nDSMs is 500 × 500 pixels. The heat maps are shown as the same size as the aerial images for demonstration reasons. This figure is best seen in color.</p> "> Figure 4
<p>The architecture of the three types of residual blocks used in the proposed models: (<b>a</b>) the typical residual block (RBLK). (<b>b</b>) The down-sampling residual block (DRBLK) uses a stride of two at both the first convolutional layer and the skip connection. (<b>c</b>) The up-sampling residual block (URBLK) uses subpixel upscaling at the first convolution and the skip connection. BN stands for batch normalization [<a href="#B55-remotesensing-13-02417" class="html-bibr">55</a>], PReLU for parametric ReLU and s is the stride of the convolutional layer.</p> "> Figure 5
<p>The architecture of the model trained with the Manchester dataset. All convolutional layers use kernel size 3 and “same” padding. BN represents a Batch Normalization layer and CNT a Concatenation layer.</p> "> Figure 6
<p>The architecture of the model trained with the DFC2018 dataset. Compared to the model trained with the Manchester dataset, the kernel sizes of certain convolutional layers are increased, and their padding is changed from “same” to “valid”, as indicated by the figure notes. Additionally, some convolutional layers are introduced at the end of the model. These modifications aim at applying a reduction factor of 10 between the input and the output of the model to match the resolution ratio between the aerial images and the nDSMs in the DFC2018 dataset. BN represents a Batch Normalization layer and CNT a Concatenation layer.</p> "> Figure 7
<p>Left: RGB images of an area in the test set of the Manchester area dataset. Middle: The ground truth nDSMs. Right: The elevation heat maps as predicted by the model. Note 1 shows cases of spurious points in the ground truth that the model correctly avoids estimating. Note 2 shows occasional inconsistencies in the dataset due to the different acquisition times of the RGB images and the LiDAR measurements. Although these inconsistencies are also evident in the training set, the model is robust to such problematic training instances. Note 3 shows cases where the model produces better-quality maps than the ground truth in terms of the surface smoothness and level of detail, as the LiDAR data contains noisy values.</p> "> Figure 8
<p>Left: RGB images from the DFC2018 test set. Middle: Ground truth nDSMs. Right: Model’s height estimations. Note 1 indicates an area that contains a group of trees and is magnified in <a href="#remotesensing-13-02417-f009" class="html-fig">Figure 9</a> to demonstrate how the model treats vegetation in the RGB images.</p> "> Figure 9
<p>Magnification of the noted region (Note 1) in <a href="#remotesensing-13-02417-f008" class="html-fig">Figure 8</a>. Left: The magnified RGB image. Middle: The ground truth nDSM. Right: Model output. The model consistently overestimates the foliage volume by filling the spaces between foliage with similar values to the neighboring estimations.</p> "> Figure 10
<p>Using a sliding window to investigate whether the model uses shadows for object height estimation. Each test case is presented in pairs of consecutive rows, the upper row showing the RGB image with the position of the sliding masking window (black square) and the lower row the prediction at the model output. The artificial shadow implied by the square black box influences the height estimation of the buildings close to the shadow by increasing their height predictions (the values on the predicted map corresponding to the buildings that are close to the shadow are seen to be brighter in the image and, thus, higher in value). The estimated heights of buildings that are not near the implied shadow are not affected. The artificial shadow causes the model to predict a higher elevation for buildings that are in the shadow’s proximity. This figure is better seen in color.</p> "> Figure 11
<p>Sample failed cases where the model misses the presence of an object completely. The cases are magnified regions from the second RGB image (second row) of <a href="#remotesensing-13-02417-f008" class="html-fig">Figure 8</a>. The top-left image shows a very high pole standing on a highway (on the left of the train wagons) with a height of 30 m (according to its LiDAR measurement). Despite the pole’s long shadow, the model does not detect it. The bottom-left magnified region contains a tall electric energy transmission tower (close and on the right of the train wagons) that is also not detected by the model.</p> ">
Abstract
:1. Introduction
2. Materials and Methods
2.1. Datasets and Data Pre-Processing
2.1.1. Manchester Area Dataset
2.1.2. IEEE GRSS Data Fusion Contest Dataset
2.2. Data Preparation
2.3. Model Description
- a typical residual block (RBLK),
- a down-sampling residual block (DRBLK) and
- an up-sampling residual block (URBLK).
2.4. Training Details
3. Results
3.1. Height Prediction for the Manchester Area Dataset
3.2. Height Prediction for the DFC2018 Dataset
3.3. Model Analysis
- Use of the up-sampling residual block (URBLK), as shown in Figure 4c, instead of the nearest-neighbor interpolation.
- Use of the down-sampling residual block (DRBLK) with strided convolutions, as shown in Figure 4b, instead of max-pooling.
- Modification of the basic U-NET scheme so that the first two concatenation layers are applied before the up-sampling steps and not after them, as originally proposed in reference [50].
- Use of “same” instead of “valid” padding in the U-NET scheme.
- Replace the ReLU activation functions with PReLUs.
3.4. Investigation of the Model’s Reliance on Shadows
3.5. Limitations
4. Discussion
5. Conclusions
Author Contributions
Funding
Acknowledgments
Conflicts of Interest
References
- Wellmann, T.; Lausch, A.; Andersson, E.; Knapp, S.; Cortinovis, C.; Jache, J.; Scheuer, S.; Kremer, P.; Mascarenhas, A.; Kraemer, R.; et al. Remote Sensing in Urban Planning: Contributions towards Ecologically Sound Policies? Landsc. Urban Plan. 2020, 204, 103921. [Google Scholar] [CrossRef]
- Bechtel, B. Recent Advances in Thermal Remote Sensing for Urban Planning and Management. In Proceedings of the Joint Urban Remote Sensing Event, JURSE 2015, Lausanne, Switzerland, 30 March–1 April 2015; IEEE: Piscataway, NJ, USA, 2015; pp. 1–4. [Google Scholar]
- Zhu, Z.; Zhou, Y.; Seto, K.C.; Stokes, E.C.; Deng, C.; Pickett, S.T.A.; Taubenböck, H. Understanding an Urbanizing Planet: Strategic Directions for Remote Sensing. Remote Sens. Environ. 2019, 228, 164–182. [Google Scholar] [CrossRef]
- Lesiv, M.; Schepaschenko, D.; Moltchanova, E.; Bun, R.; Dürauer, M.; Prishchepov, A.V.; Schierhorn, F.; Estel, S.; Kuemmerle, T.; Alcántara, C.; et al. Spatial Distribution of Arable and Abandoned Land across Former Soviet Union Countries. Sci. Data 2018, 5, 1–12. [Google Scholar] [CrossRef]
- Ma, L.; Li, M.; Blaschke, T.; Ma, X.; Tiede, D.; Cheng, L.; Chen, Z.; Chen, D. Object-Based Change Detection in Urban Areas: The Effects of Segmentation Strategy, Scale, and Feature Space on Unsupervised Methods. Remote Sens. 2016, 8, 761. [Google Scholar] [CrossRef] [Green Version]
- Muro, J.; Canty, M.; Conradsen, K.; Hüttich, C.; Nielsen, A.A.; Skriver, H.; Remy, F.; Strauch, A.; Thonfeld, F.; Menz, G. Short-Term Change Detection in Wetlands Using Sentinel-1 Time Series. Remote Sens. 2016, 8, 795. [Google Scholar] [CrossRef] [Green Version]
- Lyu, H.; Lu, H.; Mou, L. Learning a Transferable Change Rule from a Recurrent Neural Network for Land Cover Change Detection. Remote Sens. 2016, 8, 506. [Google Scholar] [CrossRef] [Green Version]
- Kaku, K. Satellite Remote Sensing for Disaster Management Support: A Holistic and Staged Approach Based on Case Studies in Sentinel Asia. Int. J. Disaster Risk Reduct. 2019, 33, 417–432. [Google Scholar] [CrossRef]
- Wing, M.G.; Burnett, J.; Sessions, J.; Brungardt, J.; Cordell, V.; Dobler, D.; Wilson, D. Eyes in the Sky: Remote Sensing Technology Development Using Small Unmanned Aircraft Systems. J. For. 2013, 111, 341–347. [Google Scholar] [CrossRef]
- Mulac, B.L. Remote Sensing Applications of Unmanned Aircraft: Challenges to Flight in United States Airspace. Geocarto Int. 2011, 26, 71–83. [Google Scholar] [CrossRef]
- Xue, F.; Lu, W.; Chen, Z.; Webster, C.J. From LiDAR Point Cloud towards Digital Twin City: Clustering City Objects Based on Gestalt Principles. ISPRS J. Photogramm. Remote Sens. 2020, 167, 418–431. [Google Scholar] [CrossRef]
- Michałowska, M.; Rapiński, J. A Review of Tree Species Classification Based on Airborne LiDAR Data and Applied Classifiers. Remote Sens. 2021, 13, 353. [Google Scholar] [CrossRef]
- Schönberger, J.L.; Frahm, J.-M. Structure-from-Motion Revisited. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA, 27–30 June 2016; IEEE Computer Society: Washington, DC, USA, 2016; pp. 4104–4113. [Google Scholar]
- Bosch, M.; Foster, K.; Christie, G.A.; Wang, S.; Hager, G.D.; Brown, M.Z. Semantic Stereo for Incidental Satellite Images. In Proceedings of the IEEE Winter Conference on Applications of Computer Vision, WACV 2019, Waikoloa Village, HI, USA, 7–11 January 2019; IEEE: Picataway, NJ, USA, 2019; pp. 1524–1532. [Google Scholar]
- Voumard, J.; Derron, M.-H.; Jaboyedoff, M.; Bornemann, P.; Malet, J.-P. Pros and Cons of Structure for Motion Embarked on a Vehicle to Survey Slopes along Transportation Lines Using 3D Georeferenced and Coloured Point Clouds. Remote Sens. 2018, 10, 1732. [Google Scholar] [CrossRef] [Green Version]
- Liu, X. Airborne LiDAR for DEM Generation: Some Critical Issues. Prog. Phys. Geogr. Earth Environ. 2008, 32, 31–49. [Google Scholar]
- Liu, C.-J.; Krylov, V.A.; Kane, P.; Kavanagh, G.; Dahyot, R. IM2ELEVATION: Building Height Estimation from Single-View Aerial Imagery. Remote Sens. 2020, 12, 2719. [Google Scholar] [CrossRef]
- He, K.; Gkioxari, G.; Dollár, P.; Girshick, R.B. Mask R-CNN. IEEE Trans. Pattern Anal. Mach. Intell. 2020, 42, 386–397. [Google Scholar] [CrossRef] [PubMed]
- Chen, L.-C.; Papandreou, G.; Schroff, F.; Adam, H. Rethinking Atrous Convolution for Semantic Image Segmentation. arXiv 2017, arXiv:1706.05587. [Google Scholar]
- Güler, R.A.; Neverova, N.; Kokkinos, I. DensePose: Dense Human Pose Estimation in the Wild. In Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, 18–22 June 2018; IEEE Computer Society: Washington, DC, USA, 2018; pp. 7297–7306. [Google Scholar]
- Szegedy, C.; Vanhoucke, V.; Ioffe, S.; Shlens, J.; Wojna, Z. Rethinking the Inception Architecture for Computer Vision. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA, 27–30 June 2016; IEEE Computer Society: Washington, DC, USA, 2016; pp. 2818–2826. [Google Scholar]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR) 2016, Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Krizhevsky, A.; Sutskever, I.; Geoffrey, E.H. ImageNet Classification with Deep Convolutional Neural Networks. In Proceedings of the Advances in Neural Information Processing Systems (NIPS), Lake Tahoe, NV, USA, 3–6 December 2012; Pereira, F., Burges, C.J.C., Bottou, L., Weinberger, K.Q., Eds.; Curran Associates, Inc.: Brooklyn, NY, USA, 2012; pp. 1097–1105. [Google Scholar]
- Tan, C.; Sun, F.; Kong, T.; Zhang, W.; Yang, C.; Liu, C. A Survey on Deep Transfer Learning. In Proceedings of the Artificial Neural Networks and Machine Learning—CANN 2018—27th International Conference on Artificial Neural Networks, Rhodes, Greece, 4–7 October 2018; Proceedings, Part III. Kurková, V., Manolopoulos, Y., Hammer, B., Iliadis, L.S., Maglogiannis, I., Eds.; Springer: New York, NY, USA, 2018; Volume 11141, pp. 270–279. [Google Scholar]
- Eigen, D.; Puhrsch, C.; Fergus, R. Depth Map Prediction from a Single Image Using a Multi-Scale Deep Network. In Proceedings of the Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, Montreal, QC, Canada, 8–13 December 2014; Ghahramani, Z., Welling, M., Cortes, C., Lawrence, N.D., Weinberger, K.Q., Eds.; Red Hook: New York, NY, USA, 2014; pp. 2366–2374. [Google Scholar]
- Laina, I.; Rupprecht, C.; Belagiannis, V.; Tombari, F.; Navab, N. Deeper Depth Prediction with Fully Convolutional Residual Networks. arXiv 2016, arXiv:1606.00373. [Google Scholar]
- Alhashim, I.; Wonka, P. High Quality Monocular Depth Estimation via Transfer Learning. arXiv 2018, arXiv:1812.11941. [Google Scholar]
- Huang, G.; Liu, Z.; Weinberger, K.Q. Densely Connected Convolutional Networks. CoRR 2016. [Google Scholar]
- Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; Fei-Fei, L. Imagenet: A Large-Scale Hierarchical Image Database. In Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition, Miami, FL, USA, 20–25 June 2009; pp. 248–255. [Google Scholar]
- Bhat, S.F.; Alhashim, I.; Wonka, P. AdaBins: Depth Estimation Using Adaptive Bins. arXiv 2020, arXiv:2011.14141. [Google Scholar]
- Geiger, A.; Lenz, P.; Urtasun, R. Are We Ready for Autonomous Driving? The KITTI Vision Benchmark Suite. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Providence, RI, USA, 16–21 June 2012. [Google Scholar]
- Nathan Silberman Derek Hoiem, P.K.; Fergus, R. Indoor Segmentation and Support Inference from RGBD Images. In Proceedings of the ECCV, Florence, Italy, 7–13 October 2012. [Google Scholar]
- Mahjourian, R.; Wicke, M.; Angelova, A. Unsupervised Learning of Depth and Ego-Motion From Monocular Video Using 3D Geometric Constraints. In Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, 18–22 June 2018; IEEE Computer Society: Washington, DC, USA, 2018; pp. 5667–5675. [Google Scholar]
- PNVR, K.; Zhou, H.; Jacobs, D. SharinGAN: Combining Synthetic and Real Data for Unsupervised Geometry Estimation. In Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020, Seattle, WA, USA, 13–19 June 2020; IEEE: Picataway, NJ, USA, 2020; pp. 13971–13980. [Google Scholar]
- Goodfellow, I.; Pouget-Abadie, J.; Mirza, M. Generative Adversarial Networks. arXiv 2014, arXiv:1406.2661. [Google Scholar] [CrossRef]
- Yu, D.; Ji, S.; Liu, J.; Wei, S. Automatic 3D Building Reconstruction from Multi-View Aerial Images with Deep Learning. ISPRS J. Photogramm. Remote Sens. 2021, 171, 155–170. [Google Scholar] [CrossRef]
- Mou, L.; Zhu, X.X. IM2HEIGHT: Height Estimation from Single Monocular Imagery via Fully Residual Convolutional-Deconvolutional Network. arXiv 2018, arXiv:1802.10249. [Google Scholar]
- Amirkolaee, H.A.; Arefi, H. Height Estimation from Single Aerial Images Using a Deep Convolutional Encoder-Decoder Network. ISPRS J. Photogramm. Remote Sens. 2019, 149, 50–66. [Google Scholar] [CrossRef]
- Srivastava, S.; Volpi, M.; Tuia, D. Joint Height Estimation and Semantic Labeling of Monocular Aerial Images with CNNS. In Proceedings of the 2017 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2017, Fort Worth, TX, USA, 23–28 July 2017; IEEE: Picataway, NJ, USA, 2017; pp. 5173–5176. [Google Scholar]
- Carvalho, M.; Le Saux, B.; Trouvé-Peloux, P.; Champagnat, F.; Almansa, A. Multitask Learning of Height and Semantics From Aerial Images. IEEE Geosci. Remote. Sens. Lett. 2020, 17, 1391–1395. [Google Scholar] [CrossRef] [Green Version]
- Ghamisi, P.; Yokoya, N. IMG2DSM: Height Simulation from Single Imagery Using Conditional Generative Adversarial Net. IEEE Geosci. Remote Sens. Lett. 2018, 15, 794–798. [Google Scholar] [CrossRef]
- Panagiotou, E.; Chochlakis, G.; Grammatikopoulos, L.; Charou, E. Generating Elevation Surface from a Single RGB Remotely Sensed Image Using Deep Learning. Remote Sens. 2020, 12, 2002. [Google Scholar] [CrossRef]
- Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; MIT Press: Cambridge, MA, USA, 2016; ISBN 9780874216561. [Google Scholar]
- Nielsen, M. Neural Networks and Deep Learning. Available online: http://neuralnetworksanddeeplearning.com/ (accessed on 24 March 2021).
- Digimap. Available online: https://digimap.edina.ac.uk/ (accessed on 25 March 2021).
- Edina. Available online: https://edina.ac.uk/ (accessed on 25 March 2021).
- Defra (Department for Environment, Food and Rural Affairs). Spatial Data. Available online: https://environment.data.gov.uk/DefraDataDownload/ (accessed on 25 March 2021).
- 2018 IEEE GRSS Data Fusion Contest. Available online: http://dase.grss-ieee.org/index.php (accessed on 24 March 2021).
- IEEE. France GRSS Chapter. Available online: https://site.ieee.org/france-grss/2018/01/16/data-fusion-contest-2018-contest-open/ (accessed on 24 March 2021).
- Ronneberger, O.; Fischer, P.; Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. In Proceedings of the International Conference of Medical Image Computing and Computer-Assisted Intervention 18 (MICCAI), Munich, Germany, 5–9 October 2015; pp. 234–241. [Google Scholar]
- Çiçek, Ö.; Abdulkadir, A.; Lienkamp, S.S.; Brox, T.; Ronneberger, O. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. In Proceedings of the Medical Image Computing and Computer-Assisted Intervention (MICCAI), Athens, Greece, 17–21 October 2016; Ourselin, S., Joskowicz, L., Sabuncu, M.R., Ünal, G.B., Wells, W., Eds.; Springer: New York, NY, USA, 2016; Volume 9901, pp. 424–432. [Google Scholar]
- Iglovikov, V.; Shvets, A. TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation. arXiv 2018, arXiv:1801.05746. [Google Scholar]
- Long, J.; Shelhamer, E.; Darrell, T. Fully Convolutional Networks for Semantic Segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, 7–12 June 2015; pp. 3431–3440. [Google Scholar]
- Shi, W.; Caballero, J.; Huszar, F.; Totz, J.; Aitken, A.P.; Bishop, R.; Rueckert, D.; Wang, Z. Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV, USA, 27–30 June 2016; IEEE Computer Society: Washington, DC, USA, 2016; pp. 1874–1883. [Google Scholar]
- Ioffe, S.; Szegedy, C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. arXiv 2015, arXiv:1502.03167. [Google Scholar]
- Kingma, D.P.; Ba, J. Adam: A Method for Stochastic Optimization. ICLR 2015, 1–15. [Google Scholar] [CrossRef]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. In Proceedings of the International Conference on Computer Vision (ICCV), Santiago, Chile, 7–13 December 2015; pp. 1026–1034. [Google Scholar]
- Van Dijk, T.; de Croon, G. How Do Neural Networks See Depth in Single Images? In Proceedings of the International Conference on Computer Vision, Seoul, Korea, 27 October–2 November 2019; IEEE: Picataway, NJ, USA, 2019; pp. 2183–2191. [Google Scholar]
- Christie, G.A.; Abujder, R.R.R.M.; Foster, K.; Hagstrom, S.; Hager, G.D.; Brown, M.Z. Learning Geocentric Object Pose in Oblique Monocular Images. In Proceedings of the Conference on Computer Vision and Pattern Recognition, CVPR 2020, Seattle, WA, USA, 13–19 June 2020; IEEE: Picataway, NJ, USA, 2020; pp. 14500–14508. [Google Scholar]
- Jones, K.L.; Poole, G.C.; O’Daniel, S.J.; Mertes, L.A.K.; Stanford, J.A. Surface Hydrology of Low-Relief Landscapes: Assessing Surface Water Flow Impedance Using LIDAR-Derived Digital Elevation Models. Remote Sens. Environ. 2008, 112, 4148–4158. [Google Scholar] [CrossRef]
- Sofia, G.; Bailly, J.; Chehata, N.; Tarolli, P.; Levavasseur, F. Comparison of Pleiades and LiDAR Digital Elevation Models for Terraces Detection in Farmlands. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2016, 9, 1567–1576. [Google Scholar] [CrossRef] [Green Version]
- Palmer, D.; Koumpli, E.; Cole, I.; Gottschalg, R.; Betts, T. A GIS-Based Method for Identification of Wide Area Rooftop Suitability for Minimum Size PV Systems Using LiDAR Data and Photogrammetry. Energies 2018, 11, 3506. [Google Scholar] [CrossRef] [Green Version]
Method | MAE(m) ↓ | RMSE(m) ↓ |
---|---|---|
Manchester Area dataset 1 | ||
IMG2nDSM * | 0.59 | 1.4 |
DFC2018 dataset 2 | ||
Carvallo et al. [40] (DSM) | 1.47 | 3.05 |
Carvallo et al. [40] (DSM + semantic) | 1.26 | 2.60 |
Liu et al. [17] | 1.19 | 2.88 |
IMG2nDSM * | 0.78 | 1.63 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Karatsiolis, S.; Kamilaris, A.; Cole, I. IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning. Remote Sens. 2021, 13, 2417. https://doi.org/10.3390/rs13122417
Karatsiolis S, Kamilaris A, Cole I. IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning. Remote Sensing. 2021; 13(12):2417. https://doi.org/10.3390/rs13122417
Chicago/Turabian StyleKaratsiolis, Savvas, Andreas Kamilaris, and Ian Cole. 2021. "IMG2nDSM: Height Estimation from Single Airborne RGB Images with Deep Learning" Remote Sensing 13, no. 12: 2417. https://doi.org/10.3390/rs13122417