A base map provides essential geospatial information for applications such as urban planning, intelligent transportation systems, and disaster management. Buildings and roads are the main ingredients of a base map and are represented by polygons. Unfortunately, manually delineating their boundaries from remote sensing data is time consuming and labour intensive. Airborne laser scanning (ALS) point clouds provide dense and accurate 3D positional information. Automatic extraction of buildings and roads from 3D point clouds is challenging because of their irregular shapes, occlusions in the data, and irregularity of ALS point clouds. This study focuses on two particular objectives: (i) accurate classification of a large volume of ALS 3D point clouds; and (ii) smooth and accurate building and road outline extraction. To achieve the classification objective, we perform point-wise deep learning to classify an ALS point cloud of a complex urban scene in Surabaya, Indonesia. The point cloud is colored by airborne orthophotos. Training data is obtained from an existing 2D topographic base map by a semi-automatic method proposed in this research. A dynamic-graph convolutional neural network is used to classify the point cloud into four classes: bare land, trees, buildings, and roads. We investigate effective input feature combinations for outdoor point cloud classification. A highly acceptable classification result of 91.8% overall accuracy is achieved when using the full combination of RGB color and LiDAR features. To address the objective of outline extraction, we propose building and road outline extraction methods that run directly on ALS point cloud data. For accurate and smooth building outline extraction, we propose two different methods. First, we develop the ordered Hough transform (OHT), which is an extension of the traditional Hough transform, by explicitly incorporating the sequence of points to form the outline. Second, we propose a new method based on Medial Axis Transform (MAT) skeletons which takes advantage of the skeleton points to detect building corners. The OHT method is resistant to noise but it requires prior knowledge on a building’s main directions. On the contrary, the MAT-based method does not require such orientation initialization but is more sensitive to noise on building edges. We compare the results of our building outline extraction methods to an existing RANSAC-based method, in terms of geometric accuracy, completeness of building corners, and computation time, and demonstrate that the MAT-based approach has the highest geometric accuracy, results in more complete building corners, and is slightly faster than other methods. For road network extraction, we develop a method based on skeletonization, which results in complete and continuous road centerlines and boundaries. In our study area, several roads are disrupted and disconnected due to trees. We design a tree-constrained approach to fill road gaps and integrate road width estimated from a medial axis algorithm. Comparison to reference data shows that the proposed method is able to extract almost all existing roads in the study area, and even detects roads that were not present in the reference due to human errors. We conclude that our object extraction methods enable a complete automatic procedure, extracting more accurate building and road outlines from ALS point cloud data. This contributes to a higher automation readiness level for a faster and cheaper base map production. @en

Classification of aerial point clouds with high accuracy is significant for many geographical applications, but not trivial as the data are massive and unstructured. In recent years, deep learning for 3D point cloud classification has been actively developed and applied, but notably for indoor scenes. In this study, we implement the point-wise deep learning method Dynamic Graph Convolutional Neural Network (DGCNN) and extend its classification application from indoor scenes to airborne point clouds. This study proposes an approach to provide cheap training samples for point-wise deep learning using an existing 2D base map. Furthermore, essential features and spatial contexts to effectively classify airborne point clouds colored by an orthophoto are also investigated, in particularly to deal with class imbalance and relief displacement in urban areas. Two airborne point cloud datasets of different areas are used: Area-1 (city of Surabaya—Indonesia) and Area-2 (cities of Utrecht and Delft—the Netherlands). Area-1 is used to investigate different input feature combinations and loss functions. The point-wise classification for four classes achieves a remarkable result with 91.8% overall accuracy when using the full combination of spectral color and LiDAR features. For Area-2, different block size settings (30, 50, and 70 m) are investigated. It is found that using an appropriate block size of, in this case, 50 m helps to improve the classification until 93% overall accuracy but does not necessarily ensure better classification results for each class. Based on the experiments on both areas, we conclude that using DGCNN with proper settings is able to provide results close to production.@en

Automatic building extraction and delineation from airborne LiDAR point cloud data of urban environments is still a challenging task due to the variety and complexity at which buildings appear. The Medial Axis Transform (MAT) is able to describe the geometric shape and topology of an object, but has never been applied for building roof outline extraction. It represents the shape of an object by its centerline, or skeleton structure instead of its boundary. Notably, end points of the MAT in principle coincide with corner points of building outlines. However, the MAT is sensitive to small boundary irregularities, which makes shape detection in airborne point clouds challenging. We propose a robust MAT-based method for detecting building corner points, which are then connected to form a building boundary polygon. First, we approximate the 2D MAT of a set of building edge points acquired by the alpha-shape algorithm to derive a so-called building roof skeleton. We then propose a hierarchical corner-aware segmentation to cluster skeleton points based on their properties which are the so-called separation angle, radius of the maximally inscribe circle, and defining edge point indices. From each segment, a corner point is then estimated by extrapolating the position of the zero radius inscribed circle based on the skeleton point positions within the segment. Our experiment uses point cloud datasets of Makassar, Indonesia and EYE-Amsterdam, The Netherlands. The average positional accuracy of the building outline results for Makassar and EYE-Amsterdam is 65 cm and 70 cm, respectively, which meet one-meter base map accuracy criteria. The results imply that skeletonization is a promising tool to extract relevant geometric information on e.g. building outlines even from far from perfect geographical point cloud data.

@en

Automation of 3D LiDAR point cloud processing is expected to increase the production rate of many applications including automatic map generation. Fast development on high-end hardware has boosted the expansion of deep learning research for 3D classification and segmentation. However, deep learning requires large amount of high quality training samples. The generation of training samples for accurate classification results, especially for airborne point cloud data, is still problematic. Moreover, which customized features should be used best for segmenting airborne point cloud data is still unclear. This paper proposes semi-automatic point cloud labelling and examines the potential of combining different tailor-made features for pointwise semantic segmentation of an airborne point cloud. We implement a Dynamic Graph CNN (DGCNN) approach to classify airborne point cloud data into four land cover classes: bare-land, trees, buildings and roads. The DGCNN architecture is chosen as this network relates two approaches, PointNet and graph CNNs, to exploit the geometric relationships between points. For experiments, we train an airborne point cloud and co-aligned orthophoto of the Surabaya city area of Indonesia to DGCNN using three different tailor-made feature combinations: points with RGB (Red, Green, Blue) color, points with original LiDAR features (Intensity, Return number, Number of returns) so-called IRN, and points with two spectral colors and Intensity (Red, Green, Intensity) so-called RGI. The overall accuracy of the testing area indicates that using RGB information gives the best segmentation results of 81.05% while IRN and RGI gives accuracy values of 76.13%, and 79.81%, respectively.

@en

Reliable and up-to-date road network information is crucial to guarantee efficient logistic distribution, emergency response, urban planning, etc. Road networks in developing urban areas tend to change rapidly. Periodic remapping is necessary to maintain the temporal quality of the road network information. Updating the road network using conventional methods can be a tedious task. This paper presents a methodology to extract road network automatically from an airborne LiDAR point cloud combined with color information from an aerial orthophoto. First, ground points are separated from non-ground points. We then classify the filtered ground points to road and non-road points using the Random Forest (RF) algorithm. Parallel thinning, method for skeletonization of the road segment, is carried out on a binary image extracted by a so-called density map of the classified road points. Finally, road centerline is obtained by our proposed topological order and regularization approach. The proposed method is tested on ISPRS benchmark data of Vaihingen - Germany. Skeleton-based road network extraction is a promising method as more than 95% roads in the study area are extracted. In the future, regularization of the skeleton to obtain smoother line representation is still an essential but challenging research.

@en

Many urban applications require building polygons as input. However, manual extraction from point cloud data is time- and labor-intensive. Hough transform is a well-known procedure to extract line features. Unfortunately, current Hough-based approaches lack flexibility to effectively extract outlines from arbitrary buildings. We found that available point order information is actually never used. Using ordered building edge points allows us to present a novel ordered points-aided Hough Transform (OHT) for extracting high quality building outlines from an airborne LiDAR point cloud. First, a Hough accumulator matrix is constructed based on a voting scheme in parametric line space (θ, r). The variance of angles in each column is used to determine dominant building directions. We propose a hierarchical filtering and clustering approach to obtain accurate line based on detected hotspots and ordered points. An Ordered Point List matrix consisting of ordered building edge points enables the detection of line segments of arbitrary direction, resulting in high-quality building roof polygons. We tested our method on three different datasets of different characteristics: one new dataset in Makassar, Indonesia, and two benchmark datasets in Vaihingen, Germany. To the best of our knowledge, our algorithm is the first Hough method that is highly adaptable since it works for buildings with edges of different lengths and arbitrary relative orientations. The results prove that our method delivers high completeness (between 90.1% and 96.4%) and correctness percentages (all over 96%). The positional accuracy of the building corners is between 0.2-0.57 m RMSE. The quality rate (89.6%) for the Vaihingen-B benchmark outperforms all existing state of the art methods. Other solutions for the challenging Vaihingen-A dataset are not yet available, while we achieve a quality score of 93.2%. Results with arbitrary directions are demonstrated on the complex buildings around the EYE museum in Amsterdam.@en

Change detection is an essential step to locate the area where an old model should be updated. With high density and accuracy, LiDAR data is often used to create a 3D city model. However, updating LiDAR data at state or nation level often takes years. Very high resolution (VHR) images with high updating rate is therefore an option for change detection. This paper provides a novel and efficient approach to derive pixel-based building change detection between past LiDAR and new VHR images. The proposed approach aims notably at reducing false alarms of changes near edges. For this purpose, LiDAR data is used to supervise the process of finding stereo pairs and derive the changes directly. This paper proposes to derive three possible heights (so three DSMs) by exploiting planar segments from LiDAR data. Near edges, the up to three possible heights are transformed into discrete disparities. A optimal disparity is selected from a reasonable and computational efficient range centered on them. If the optimal disparity is selected, but still the stereo pair found is wrong, a change has been found. A Markov random field (MRF) with built-in edge awareness from images is designed to find optimal disparity. By segmenting the pixels into plane and edge segments, the global optimization problem is split into many local ones which makes the optimization very efficient. Using an optimization and a consecutive occlusion consistency check, the changes are derived from stereo pairs having high color difference. The algorithm is tested to find changes in an urban areas in the city of Amersfoort, the Netherlands. The two different test cases show that the algorithm is indeed efficient. The optimized disparity images have sharp edges along those of images and false alarms of changes near or on edges and occlusions are largely reduced.

@en

Various kinds of urban applications require true orthophotos. True orthophoto generation requires a DSM (Digital Surface Model) to project the photo orthogonally and minimize geometric distortion due to topographic variance. DSMs are often generated from airborne laser scan data. In urban scenes, DSM data may fail to deliver sharp and straight building roof edges. This will affect the quality of the resulting orthophotos. Therefore, it is necessary to incorporate good quality building outlines as breaklines during DSM interpolation. This study proposes a data-driven approach to construct building roof outlines from LiDAR point clouds by a workflow consisting of the following steps: given roof segments, roof boundary points are extracted using a concave hull algorithm. Straight edges may be difficult to find in complex roof configurations. Therefore, two ingredients are combined. First, RanSAC corner point preselection, and second, DBSCAN-based clustering of edge points. The method is demonstrated on an area of ±1.2 km² containing 42 buildings of different characteristics. A quality assessment shows that the proposed method is able to deliver 92% of building lines with acceptable geometric accuracy in comparison to a building line in the base map.

@en

Semantic segmentation, especially for buildings, from the very high resolution (VHR) airborne images is an important task in urban mapping applications. Nowadays, the deep learning has significantly improved and applied in computer vision applications. Fully Convolutional Networks (FCN) is one of the tops voted method due to their good performance and high computational efficiency. However, the state-of-art results of deep nets depend on the training on large-scale benchmark datasets. Unfortunately, the benchmarks of VHR images are limited and have less generalization capability to another area of interest. As existing high precision base maps are easily available and objects are not changed dramatically in an urban area, the map information can be used to label images for training samples. Apart from object changes between maps and images due to time differences, the maps often cannot perfectly match with images. In this study, the main mislabeling sources are considered and addressed by utilizing stereo images, such as relief displacement, different representation between the base map and the image, and occlusion areas in the image. These free training samples are then fed to a pre-trained FCN. To find the better result, we applied fine-tuning with different learning rates and freezing different layers. We further improved the results by introducing atrous convolution. By using free training samples, we achieve a promising building classification with 85.6% overall accuracy and 83.77% F1 score, while the result from ISPRS benchmark by using manual labels has 92.02% overall accuracy and 84.06% F1 score, due to the building complexities in our study area.

@en

LiDAR data acquisition is recognized as one of the fastest solutions to provide basis data for large-scale topographical base maps worldwide. Automatic LiDAR processing is believed one possible scheme to accelerate the large-scale topographic base map provision by the Geospatial Information Agency in Indonesia. As a progressive advanced technology, Geographic Information System (GIS) open possibilities to deal with geospatial data automatic processing and analyses. Considering further needs of spatial data sharing and integration, the one stop processing of LiDAR data in a GIS environment is considered a powerful and efficient approach for the base map provision. The quality of the automated topographic base map is assessed and analysed based on its completeness, correctness, quality, and the confusion matrix.

@en

The integration of computer vision and photogrammetry to generate three-dimensional (3D) information from images has contributed to a wider use of point clouds, for mapping purposes. Large-scale topographic map production requires 3D data with high precision and accuracy to represent the real conditions of the earth surface. Apart from LiDAR point clouds, the image-based matching is also believed to have the ability to generate reliable and detailed point clouds from multiple-view images. In order to examine and analyze possible fusion of LiDAR and image-based matching for large-scale detailed mapping purposes, point clouds are generated by Semi Global Matching (SGM) and by Structure from Motion (SfM). In order to conduct comprehensive and fair comparison, this study uses aerial photos and LiDAR data that were acquired at the same time. Qualitative and quantitative assessments have been applied to evaluate LiDAR and image-matching point clouds data in terms of visualization, geometric accuracy, and classification result. The comparison results conclude that LiDAR is the best data for large-scale mapping.

@en