Fig 4 - uploaded by Caroline Gevaert
Content may be subject to copyright.
Images of the Kigali (a), Dar es Salaam (b), and Lombardia (c) datasets, and their respective DSMs (d-f) and manual reference data (g-i). 

Images of the Kigali (a), Dar es Salaam (b), and Lombardia (c) datasets, and their respective DSMs (d-f) and manual reference data (g-i). 

Source publication
Article
Full-text available
Existing algorithms for Digital Terrain Model (DTM) extraction still face difficulties due to data outliers and geometric ambiguities in the scene such as contiguous off-ground areas or sloped environments. We postulate that in such challenging cases, the radiometric information contained in aerial imagery may be leveraged to distinguish between gr...

Contexts in source publication

Context 1
... first dataset consists of UAV imagery collected over an informal settlement in Kigali, Rwanda (Fig. 4a,d). Images were collected with a DJI Phantom 2 Vision+ quadcopter and processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 3 cm. A subset of 5000 × 5000 pixels (150 × 150 m) was selected which contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The ...
Context 2
... contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The terrain of the lower part of the image contains steep slopes, making it a challenging scene for DTM extraction algorithms. More information regarding the UAV data col- lection and processing can be found in Gevaert et al., (2017). The re- ference data (Fig. 4g) was manually created by visual ...
Context 3
... second dataset consists of UAV imagery over Dar es Salaam, Tanzania (Fig. 4b,e). The images were collected in 2015 with a SenseFly eBee mounted with a 14 MP Canon Powershot RGB camera in the context of a World Bank project (Dar Ramani Huria 2 ). These images were processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 5 cm. A subset of 6000 × 6000 pixels (300 × 300 m) ...
Context 4
... again covers an informal settlement. Although the area is not as steeply sloped as in Kigali, the area also challenging due to the presence of contiguous off-ground areas and spectral similarity between the ground and off-ground objects. Reference data for the ground and off-ground object classes was again manually digitized over the orthomosaic (Fig. ...
Context 5
... Italy with a Vexcel UltraCam Xp on May 29, 2015. The aerial images were processed to obtain an orthomosaic and DSM with a Ground Sampling Distance (GSD) of 20 cm. A subset of 5000 × 5000 pixels (1000 × 1000 m) was selected for the experimental analyses. The area consists of a residential area, river, dense forests, agricultural fields and a dike (Fig. 4c,f). A DTM of this area was obtained by the Compagnia Generale Ripre- seaeree (CGR S.p.A.) by manually editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and ...
Context 6
... editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and pixels where they were equal as ground. Pixels where the difference was between 0 and 0.5 m were left unlabelled (Fig. ...
Context 7
... first dataset consists of UAV imagery collected over an informal settlement in Kigali, Rwanda (Fig. 4a,d). Images were collected with a DJI Phantom 2 Vision+ quadcopter and processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 3 cm. A subset of 5000 × 5000 pixels (150 × 150 m) was selected which contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The terrain of the lower part of the image contains steep slopes, making it a challenging scene for DTM extraction algorithms. More information regarding the UAV data col- lection and processing can be found in Gevaert et al., (2017). The re- ference data (Fig. 4g) was manually created by visual ...
Context 8
... first dataset consists of UAV imagery collected over an informal settlement in Kigali, Rwanda (Fig. 4a,d). Images were collected with a DJI Phantom 2 Vision+ quadcopter and processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 3 cm. A subset of 5000 × 5000 pixels (150 × 150 m) was selected which contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The terrain of the lower part of the image contains steep slopes, making it a challenging scene for DTM extraction algorithms. More information regarding the UAV data col- lection and processing can be found in Gevaert et al., (2017). The re- ference data (Fig. 4g) was manually created by visual ...
Context 9
... second dataset consists of UAV imagery over Dar es Salaam, Tanzania (Fig. 4b,e). The images were collected in 2015 with a SenseFly eBee mounted with a 14 MP Canon Powershot RGB camera in the context of a World Bank project (Dar Ramani Huria 2 ). These images were processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 5 cm. A subset of 6000 × 6000 pixels (300 × 300 m) was selected for the current analysis. The area again covers an informal settlement. Although the area is not as steeply sloped as in Kigali, the area also challenging due to the presence of contiguous off-ground areas and spectral similarity between the ground and off-ground objects. Reference data for the ground and off-ground object classes was again manually digitized over the orthomosaic (Fig. ...
Context 10
... second dataset consists of UAV imagery over Dar es Salaam, Tanzania (Fig. 4b,e). The images were collected in 2015 with a SenseFly eBee mounted with a 14 MP Canon Powershot RGB camera in the context of a World Bank project (Dar Ramani Huria 2 ). These images were processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 5 cm. A subset of 6000 × 6000 pixels (300 × 300 m) was selected for the current analysis. The area again covers an informal settlement. Although the area is not as steeply sloped as in Kigali, the area also challenging due to the presence of contiguous off-ground areas and spectral similarity between the ground and off-ground objects. Reference data for the ground and off-ground object classes was again manually digitized over the orthomosaic (Fig. ...
Context 11
... third dataset was obtained over Lombardia, Italy with a Vexcel UltraCam Xp on May 29, 2015. The aerial images were processed to obtain an orthomosaic and DSM with a Ground Sampling Distance (GSD) of 20 cm. A subset of 5000 × 5000 pixels (1000 × 1000 m) was selected for the experimental analyses. The area consists of a residential area, river, dense forests, agricultural fields and a dike (Fig. 4c,f). A DTM of this area was obtained by the Compagnia Generale Ripre- seaeree (CGR S.p.A.) by manually editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and pixels where they were equal as ground. Pixels where the difference was between 0 and 0.5 m were left unlabelled (Fig. ...
Context 12
... third dataset was obtained over Lombardia, Italy with a Vexcel UltraCam Xp on May 29, 2015. The aerial images were processed to obtain an orthomosaic and DSM with a Ground Sampling Distance (GSD) of 20 cm. A subset of 5000 × 5000 pixels (1000 × 1000 m) was selected for the experimental analyses. The area consists of a residential area, river, dense forests, agricultural fields and a dike (Fig. 4c,f). A DTM of this area was obtained by the Compagnia Generale Ripre- seaeree (CGR S.p.A.) by manually editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and pixels where they were equal as ground. Pixels where the difference was between 0 and 0.5 m were left unlabelled (Fig. ...
Context 13
... first dataset consists of UAV imagery collected over an informal settlement in Kigali, Rwanda (Fig. 4a,d). Images were collected with a DJI Phantom 2 Vision+ quadcopter and processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 3 cm. A subset of 5000 × 5000 pixels (150 × 150 m) was selected which contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The terrain of the lower part of the image contains steep slopes, making it a challenging scene for DTM extraction algorithms. More information regarding the UAV data col- lection and processing can be found in Gevaert et al., (2017). The re- ference data (Fig. 4g) was manually created by visual ...
Context 14
... first dataset consists of UAV imagery collected over an informal settlement in Kigali, Rwanda (Fig. 4a,d). Images were collected with a DJI Phantom 2 Vision+ quadcopter and processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 3 cm. A subset of 5000 × 5000 pixels (150 × 150 m) was selected which contains densely grouped buildings separated by narrow foot- paths which are often shadowed. The terrain of the lower part of the image contains steep slopes, making it a challenging scene for DTM extraction algorithms. More information regarding the UAV data col- lection and processing can be found in Gevaert et al., (2017). The re- ference data (Fig. 4g) was manually created by visual ...
Context 15
... second dataset consists of UAV imagery over Dar es Salaam, Tanzania (Fig. 4b,e). The images were collected in 2015 with a SenseFly eBee mounted with a 14 MP Canon Powershot RGB camera in the context of a World Bank project (Dar Ramani Huria 2 ). These images were processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 5 cm. A subset of 6000 × 6000 pixels (300 × 300 m) was selected for the current analysis. The area again covers an informal settlement. Although the area is not as steeply sloped as in Kigali, the area also challenging due to the presence of contiguous off-ground areas and spectral similarity between the ground and off-ground objects. Reference data for the ground and off-ground object classes was again manually digitized over the orthomosaic (Fig. ...
Context 16
... second dataset consists of UAV imagery over Dar es Salaam, Tanzania (Fig. 4b,e). The images were collected in 2015 with a SenseFly eBee mounted with a 14 MP Canon Powershot RGB camera in the context of a World Bank project (Dar Ramani Huria 2 ). These images were processed with Pix4Dmapper to obtain a DSM and true-colour orthomosaic with a spatial resolution of 5 cm. A subset of 6000 × 6000 pixels (300 × 300 m) was selected for the current analysis. The area again covers an informal settlement. Although the area is not as steeply sloped as in Kigali, the area also challenging due to the presence of contiguous off-ground areas and spectral similarity between the ground and off-ground objects. Reference data for the ground and off-ground object classes was again manually digitized over the orthomosaic (Fig. ...
Context 17
... third dataset was obtained over Lombardia, Italy with a Vexcel UltraCam Xp on May 29, 2015. The aerial images were processed to obtain an orthomosaic and DSM with a Ground Sampling Distance (GSD) of 20 cm. A subset of 5000 × 5000 pixels (1000 × 1000 m) was selected for the experimental analyses. The area consists of a residential area, river, dense forests, agricultural fields and a dike (Fig. 4c,f). A DTM of this area was obtained by the Compagnia Generale Ripre- seaeree (CGR S.p.A.) by manually editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and pixels where they were equal as ground. Pixels where the difference was between 0 and 0.5 m were left unlabelled (Fig. ...
Context 18
... third dataset was obtained over Lombardia, Italy with a Vexcel UltraCam Xp on May 29, 2015. The aerial images were processed to obtain an orthomosaic and DSM with a Ground Sampling Distance (GSD) of 20 cm. A subset of 5000 × 5000 pixels (1000 × 1000 m) was selected for the experimental analyses. The area consists of a residential area, river, dense forests, agricultural fields and a dike (Fig. 4c,f). A DTM of this area was obtained by the Compagnia Generale Ripre- seaeree (CGR S.p.A.) by manually editing the DSM. Therefore, the re- ference data for the classification part of the experimental analyses was determined by classifying all pixels where the difference between the DSM and DTM was greater than 50 cm as off-ground, and pixels where they were equal as ground. Pixels where the difference was between 0 and 0.5 m were left unlabelled (Fig. ...

Similar publications

Article
Full-text available
The registration of multi-temporal remote sensing images with abundant information and complex changes is an important preprocessing step for subsequent applications. This paper presents a novel two-stage deep learning registration method based on sub-image matching. Unlike the conventional registration framework, the proposed network learns the ma...
Article
Full-text available
Image registration is a viable task in the field of computer vision with many applications. When images are captured under different spectrum conditions, a challenge is imposed on the task of registration. Researchers carefully handcraft a local module insensitive to illumination changes across cross-spectral image pairs to tackle this challenge. W...

Citations

... Zhang et al. 2013). Recently, deep-learning methods have been developed for point-cloud filtering (Hu and Yuan 2016;Gevaert et al. 2018;Rizaldy et al. 2018;Jin et al. 2020). Although these machine-learning methods have good outcomes in lidar point clouds, they are not generalizable for photogrammetric point clouds. ...
Article
Full-text available
Ground-point filtering from point-cloud data is an important process in remote sensing and the photogrammetric map-production line, especially in generating digital elevation models from airborne lidar and aerial photogrammetric point-cloud data. In this article, a new and simple boundary-based method is proposed for ground-point filtering from the photogrammetric point-cloud data. The proposed method uses the local height difference to extract the boundaries of objects. Then the extracted boundary points are traced to generate polygons around the borders of any objects on the ground. Finally, the points located inside these polygons, which are classified as non-ground points, are filtered. The experimental results on the photogrammetric point cloud show that the proposed method can adapt to complex environments. The total error of the proposed method is about 8.96%, which is promising in these challenging data sets. Moreover, the proposed method is compared with cloth simulation filtering, multi-scale curvature classification, and gLiDAR methods and gives better results.
... Considering that most DTM generating algorithms share the common goal of discriminating between smooth and protruding shapes, algorithms can be classified according to how to represent 3-dimensional coordinate measurements. Point cloud (Bartels and Wei, 2010;Bartels et al., 2006;Sithole and Vosselman, 2005;Vosselman, 2000;Zhang et al., 2016), triangulated irregular network (TIN) (Axelsson, 2000;Sohn and Dowman, 2002;Zhang and Lin, 2013), and image grid (Amirkolaee et al., 2022;Chen et al., 2012;Gevaert et al., 2018;Hu and Yuan, 2016;Lohmann et al., 2000;Mongus andŽalik, 2013;Wack and Wimmer, 2002;Zhang et al., 2003) are the three most common representations of 3-dimensional coordinate measurements. ...
... Alternative recent methods for generating DTM algorithms include deep learning-based methods (Amirkolaee et al., 2022;Gevaert et al., 2018;Hu and Yuan, 2016) and a cloth simulation-based method (Zhang et al., 2016). Deep learning-based algorithms usually regard DSM as an image and try to extract non-ground pixels similar to the common approach for computervision tasks, namely, semantic segmentation or object detection. ...
... Also, in general, external sources for water mapping are readily available due to lots of accumulated remotely sensed imagery (Huang et al., 2018). Perhaps these are the reasons why most DTM generation algorithms have ignored water body mapping (Hu and Yuan, 2016;Gevaert et al., 2018;Amirkolaee et al., 2022) even if subsequent analyses based on DTM often require a water map. However, the use of external data sources can lead to errors due to registration issues or mismatches in spatial and temporal resolution with the LiDAR data. ...
Preprint
Full-text available
The digital terrain model (DTM) is fundamental geospatial data for various studies in urban, environmental, and Earth science. The reliability of the results obtained from such studies can be considerably affected by the errors and uncertainties of the underlying DTM. Numerous algorithms have been developed to mitigate the errors and uncertainties of DTM. However, most algorithms involve tricky parameter selection and complicated procedures that make the algorithm's decision rule obscure, so it is often difficult to explain and predict the errors and uncertainties of the resulting DTM. Also, previous algorithms often consider the local neighborhood of each point for distinguishing non-ground objects, which limits both search radius and contextual understanding and can be susceptible to errors particularly if point density varies. This study presents an open-source DTM generation algorithm for airborne LiDAR data that can consider beyond the local neighborhood and whose results are easily explainable, predictable, and reliable. The key assumption of the algorithm is that grounds are smoothly connected while non-grounds are surrounded by areas having sharp elevation changes. The robustness and uniqueness of the proposed algorithm were evaluated in geographically complex environments through tiling evaluation compared to other state-of-the-art algorithms.
... DTM extraction can be achieved from satellite or aerial images [6], [7], DSMs [8], or point clouds [9], [10]. Among the possibilities, airborne laser scanning (ALS) point clouds, with rich geometry information (c.f . ...
... vegetation, construction) in geospatial data, efforts have been focused on exploring ground filtering. One strategy is to rely on terrain elevation and slope to establish rules that separate objects from ground [6], [9], [25], [26], under the assumption of gradual relief and small land covers with respect to land scales. Classification is performed by thresholding over pre-defined values either by chosen manually or from local adaptive filtering based on terrain slope [25]. ...
... In recent years, the paradigm is shifted toward more of learning-based methods [6], [14], [18], [19]. Such methods find their emergence from the breakthroughs of deep networks [15], [16], [32], [33], which, inspired by human neural systems, model and learn parametric nonlinear functions defined by sequentially stacked layers (hence deep network) from massive amount of data. ...
Preprint
Full-text available
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.
... Hu and Yuan (2016) employed a deep convolutional neural network (CNN) to classify the points into ground and non-ground classes (Hu and Yuan, 2016). Gevaert et al. (2018) applied a morphological filter to get candidate ground and non-ground samples and then the acquired samples are used to train a fully convolutional network (FCN) (Gevaert et al., 2018). Diakogiannis et al., (2020) proposed a UNet encoder-decoder backbone that contains the residual connections, atrous convolutions, pyramid scene parsing pooling, and multi-tasking inference for image classification (Diakogiannis et al., 2020). ...
... Hu and Yuan (2016) employed a deep convolutional neural network (CNN) to classify the points into ground and non-ground classes (Hu and Yuan, 2016). Gevaert et al. (2018) applied a morphological filter to get candidate ground and non-ground samples and then the acquired samples are used to train a fully convolutional network (FCN) (Gevaert et al., 2018). Diakogiannis et al., (2020) proposed a UNet encoder-decoder backbone that contains the residual connections, atrous convolutions, pyramid scene parsing pooling, and multi-tasking inference for image classification (Diakogiannis et al., 2020). ...
... All existed methods to extract DTM from DSM using deep learning concepts are based on classification (Gevaert et al., 2018;Hu and Yuan, 2016). In these methods, firstly, a network is trained to classify the input data into the ground and non-ground pixels, and then the non-ground pixels are eliminated. ...
Article
Full-text available
Extraction of digital terrain model (DTM) from Digital Surface Model (DSM) still faces many problems in a complex scene with geometric ambiguities such as steep slope forested environments and contiguous non-ground regions. In this paper, an approach based on deep learning is proposed to generate DTM directly from DSM without applying filtering methods for eliminating non-ground pixels. In this regard, first, in the preprocessing step, the data is prepared for entering into the proposed deep network. Then, a hybrid deep convolutional neural network (HDCNN) is proposed which is a combination of the U-net architecture and residual networks. In this network, effective features are generate4d in different scales during the downsampling process from the input DSM and the DTM is extracted during the upsampling process. To rectify the results, a multi-scale fusion strategy is proposed to produce the final DTM by fusing the generated DTMs at different scales and with different spatial shifts. The performance of the proposed approach is analyzed by implementing four different evaluation scenarios in five different datasets. The evaluation results demonstrated significant performance and high generalizability of the proposed approach. The proposed network also outperforms the deep learning-based filtering methods and two reference DTM extraction algorithms especially in challenging regions.
... e extremely high spatial resolution of 2D images combined with 3D geometric information allows us not just to recognize a large and detailed set of thematic classes, but also to precisely characterize the objects in the area under investigation according to their material and geometry. In the context of urban studies, UAV data can be used to map buildings (Gevaert et al., 2018a), roads (Zhou et al., 2017), monuments (Fiorillo et al., 2013) as well as to detect damages a er catastrophic events (Nex & Remondino, 2014) (e.g., earthquakes, ooding). In vegetation related studies, UAV data can provide estimations of biophysical parameters or detect early signs of plant stress or disease (Nex & Remondino, 2014). ...
Book
Full-text available
You can download the e-book here: https://files.wbg-wissenverbindet.de/Files/Article/ARTK_ZOA_1028514_0002.pdf
... The ground, building, low vegetation, medium vegetation, high vegetation, low point, air point, and default classes were obtained based on the standard point classes of the American Society for Photogrammetry and Remote Sensing (ASPRS) [65]. In the rule-based classification algorithms, the information about the terrain surface is converted to a set of rules; thus each terrain class has its own characteristic rules, and the classification is carried out based on these rules [66,67]. In this study, each individual LiDAR point was classified into the appropriate classes with point-based rules using spatial features. ...
Article
Full-text available
Trees are the key components of urban vegetation in cities. The timely and accurate identification of existing urban tree species with their location is the most important task for improving air, water, and land quality; reducing carbon accumulation; mitigating urban heat island effects; and protecting soil and water balance. Light detection and ranging (LiDAR) is frequently used for extracting high-resolution structural information regarding tree objects. LiDAR systems are a cost-effective alternative to the traditional ways of identifying tree species, such as field surveys and aerial photograph interpretation. The aim of this work was to assess the usage of machine learning algorithms for classifying the deciduous (broadleaf) and coniferous tree species from 3D raw LiDAR data on the Davutpasa Campus of Yildiz Technical University, Istanbul, Turkey. First, ground, building, and low, medium, and high vegetation classes were acquired from raw LiDAR data using a hierarchical-rule-based classification method. Next, individual tree crowns were segmented using a mean shift clustering algorithm from high vegetation points. A total of 25 spatial- and intensity-based features were utilized for support vector machine (SVM), random forest (RF), and multi-layer perceptron (MLP) classifiers to discriminate deciduous and coniferous tree species in the urban area. The machine learning-based classification’s overall accuracies were 80%, 83.75%, and 73.75% for the SVM, RF, and MLP classifiers, respectively, in split 70/30 (training/testing). The SVM and RF algorithms generally gave better classification results than the MLP algorithm for identifying the urban tree species.
... On the other hand, deep-learning based algorithms have undergone a rapid improvement in accuracy in wide range of image-based analysis including supervised image classification tasks (Gevaert et al. 2018), dense matching of MVS imagery (Gu et al. 2020, Liu andJi 2020) and feature extraction from DSMs (Yu et al. 2021). As an example a in a recent work implemented by Yu et al. 2021 a multiple scale pyramid matching pipeline based on a convolutional neural network to extract DSMs automatically from aerial imagery. ...
... This additional information contributes to characterize spatial features of target objects in the 3D environment. Promising results have been shown in Gevaert et al. (2018) where 2.5D topographic and 3D geometric features are integrated into a classical machine learning workflow to improve accuracy in complex classification problems. This integration of 2D and 3D information could lead to improved performance in many other applications in the future. ...
... Given the concurrent work of different domains and the optimization of dedicated hardware to run them onboard and in real-time, this trend will grow in the future. This process will be facilitated by the availability of more complete training datasets (including synthetic data as well) or by emerging approaches such as self-supervised learning, reinforcement learning or the automatic generation of pseudo-labels exploiting already existing data (Gevaert et al., 2018). Questions regarding the transferability of these models to heterogenous and unexpected operational conditions will need to be addressed with rigorous and innovative strategies. ...
Article
Full-text available
The use of Unmanned Aerial Vehicles (UAVs) has surged in the last two decades, making them popular instruments for a wide range of applications, and leading to a remarkable number of scientific contributions in geoscience, remote sensing and engineering. However, the development of best practices for high quality of UAV mapping are often overlooked representing a drawback for their wider adoption. UAV solutions then require an inter-disciplinary research, integrating different expertise and combining several hardware and software components on the same platform. Despite the high number of peer-reviewed papers on UAVs, little attention has been given to the interaction between research topics from different domains (such as robotics and computer vision) that impact the use of UAV in remote sensing. The aim of this paper is to (i) review best practices for the use of UAVs for remote sensing and mapping applications and (ii) report on current trends-including adjacent domains-for UAV use and discuss their future impact in photogrammetry and remote sensing. Hardware developments , navigation and acquisition strategies, and emerging solutions for data processing in innovative applications are considered in this analysis. As the number and the heterogeneity of debated topics are large, the paper is organized according to very specific questions considered most relevant by the authors.
... The discriminant function assumes or models the local terrain attributes such as elevation, slope, curvature, and surface homogeneity. LiDAR data can be filtered by methods based on slope [18]- [20], morphology [21]- [24], surface [25], [26], statistics [27], linear prediction [28], [29], triangulated irregular network (TIN) [30]- [32], interpolation [31], [32], segmentation/cluster [35], [36], and machine learning algorithm [37]- [39], or by intelligent algorithms and hybrid methods [38], [39]. The purpose of the interpolation process is to reconstruct the surface Yu Tian, Zhengfu Bian, Shaogang Lei, Chuning Ji, Yibo Zhao, Shubi Zhang, Lei Duan, Vladimí r Sedlá k A process-oriented method for rapid acquisition of canopy height model from RGB point cloud in Semi-arid region using a discontinuous sequence of elevations. ...
Article
Full-text available
This study addressed the problems of difficult parameter selection of ground point filtering algorithm, slow speed, and susceptibility to wrong classification points of traditional interpolation algorithm in extracting canopy height model (CHM) from dense point cloud produced by structure-from-motion (SfM). Finding the true ground is vital for canopy height estimation using UAV RGB point cloud, especially for semi-arid area where cloud points of short vegetation like shrubs and dwarf trees generally mixed up with the ground points. This paper proposes a ground point extraction strategy that combines the advantages of structural and spectral filtering. Specifically, spectral filtering simplifies the threshold selection for structural filtering, while structural filtering removes the outliers from spectral filtering. Considering the misclassified points in the ground filtering algorithm, a fast, nonparametric ground point interpolation algorithm was used to suppress the wrong classification points. This novel algorithm is based on the basic idea of least square quadratic fitting and prediction of terrain profile (PF). The quantitative results show that compared to Inverse Distance Weighting (IDW), Radial Basis Function (RBF), and Ordinary Kriging (OK), it takes less time (PF: 122s, IDW: 518s, RBF: 1374s, OK: 1129s), and has lower RMSE (PF: 0.301 m, IDW: 0.549 m, RBF: 0.903m, OK: 0.427m).
... Multi-directional and Slope Dependent (MSD) [16], or even deep learning methods e.g. [7]. These approaches cannot overcome the interpolation effects without negative impacts on the final modelling results. ...