Billions Of Points Research Articles

A Scientific Investigation of the Shangfang Mountain Yunshui Cave in Beijing Based on LiDAR Technology

The Yunshui Cave in Shangfang Mountain, Beijing, is a famous high-altitude karst cave in northern China. As the third scientific survey of Yunshui Cave in history, this is the first time to use the latest LiDAR technology to carry out a related detection survey. Traditional cave measurement methods are limited by natural conditions and make it difficult to reach the destination. Traditional methods mainly rely on experience and obtain data with strong subjectivity, making it difficult to conduct quantitative research and obtain reproducible results in the current information era. Applying LiDAR technology to cave measurement can obtain comprehensive and accurate digital measurement results within the same survey time and reveal many richer and more accurate features of Yunshui Cave. The obtained digital measurement results can be used for 3D modeling as well as provide a large amount of accurate basic data and preliminary materials for subsequent geological, environmental, and archaeological investigation and analysis, as well as cultural and tourism resource development. The rapid geological survey of Shangfang Mountain Yunshui Cave using LiDAR technology shows that LiDAR cave geological survey technology can achieve real-time collection of centimeter-level accuracy and generate billions of points of cloud data, greatly improving survey efficiency and accuracy. At the same time, digital survey results can be obtained. Through modeling and GIS technology, all on-site survey details can be easily moved back to the laboratory for real-scene reproduction, network sharing, and dissemination. This study provides a foundation for future explorations of the Yunshui cave and highlights the potential for LiDAR techniques to enhance our understanding of complex geological structures such as caves.

VxH: A Systematic Determination of Efficient Hierarchical Voxel Structures

Three-dimensional (3D) maps with many millions to billions of points are now used in an increasing number of applications, with processing rates in the hundreds of thousands to millions of points per second. In mobile applications, power and energy consumption for managing such data and extracting useful information thereof are critical concerns. We have developed structures and methodologies with the purpose of minimizing memory usage and associated energy consumption for indexing and serialization of voxelized point-clouds. The primary source of points in our case is airborne laser scanning, but our methodology is not restricted to only this setting. Our emulated results show a memory usage reduction factor of roughly up to 200× that of Octree/Octomap and a file size reduction factor of up to 1.65× compared with the predominating compression scheme for airborne Lidar data, LASzip. In addition, our structures enable significantly more efficient processing since they are included in a hierarchical structure that captures geometric aspects.

Vo-Norvana : Versatile Framework for Efficient Segmentation of Large Point Cloud Data Sets

Dense three-dimensional (3D) point clouds collected from rapidly evolving data acquisition techniques such as light detection and ranging (lidar) and structure from motion (SfM) multiview stereo (MVS) photogrammetry contain detailed geometric information of a scene suitable for a wide variety of applications. Among the many processes within a typical point cloud processing workflow, segmentation is often a crucial step to group points with similar attributes to support more advanced modeling and analysis. Segmenting large point cloud data sets (i.e., hundreds of millions to billions of points) can be extremely time consuming and tedious with current tools, which primarily rely on significant manual effort. While many automated methods have been proposed, the practicality, scalability, and versatility of these approaches remain a bottleneck stifling processing of large data sets. To overcome these challenges, this paper introduces a novel, generalized segmentation framework called Vo-Norvana, which incorporates a new voxelization technique, a normal variation analysis considering the positioning uncertainty of the point cloud, and a custom region growing process for clustering. The proposed framework was tested with several large-volume data sets collected in diverse scene types using several data acquisition platforms including terrestrial lidar, mobile lidar, airborne lidar, and drone-based SfM-MVS photogrammetry. In evaluating the accuracy of models generated from Vo-Norvana against manual segmentation, the average error of the position, orientation, and dimensions are 2.7 mm, 0.083°, and 0.9 mm, respectively. Over 0.2 million points per second and 36 thousand voxels per second can be achieved when segmenting an airborne lidar data set containing over 639 million points to about 1 million segments.

SHREC 2023: Point cloud change detection for city scenes

Localization and navigation are the two most important tasks for mobile robots, which require an up-to-date and accurate map. However, to detect map changes from crowdsourced data is a challenging task, especially from billions of points collected by 3D acquisition devices. Collecting 3D data often requires expensive data acquisition equipment and there are limited data sources to evaluate point cloud change detection. To address these issues, in this Shape Retrieval Challenge (SHREC) track, we provide a city-scene dataset with real and synthesized data to detect 3D point cloud change. The dataset consists of 866 pairs of object changes from 78 city-scene 3D point clouds collected by LiDAR and 845 pairs of object changes from 100 city-scene 3D point clouds generated by a high-fidelity simulator.We compare three methods on this benchmark. Evaluation results show that data-driven methods are the current trend in 3D point cloud change detection. Besides, the siamese network architecture is helpful to detect changes in our dataset. We hope this benchmark and comparative evaluation results will further enrich and boost the research of point cloud change detection and its applications.

Dark cities or cities of light? – sunlight amenity preservation at whole-city scale using a spatio-temporal decision support approach

Public open spaces are fundamentally important for the health and well-being of citizens in densely populated cities. If not carefully planned, high-density urban development can overshadow adjacent open spaces, resulting in poor quality, dark and oppressive winter conditions. Current planning control approaches for protecting light amenity in cities are often limited to simple overshadowing impact diagrams (e.g. shadows cast on the equinox at 9 a.m., 12 p.m. and 3 p.m.). In cities transitioning from low to higher density, comprising more complex urban forms and more extreme seasonal light amenity dynamics, these static approaches are insufficient. This paper outlines the development of a spatio-temporal design decision support system for analysing and protecting the light amenity of public open spaces, applied to a capital city in Australia. The system described has two parts: Firstly, to assess the overshadowing of existing public open spaces and identify those in need of protection (Part A), and secondly to generate planning restrictions to protect designated open spaces from future deprivation of light (Part B). For Part A, we use a graphics processing unit accelerated aggregate-shadow (15 minute increments) calculation applied to a detailed city-wide 3D model generated from billions of points of aerial survey (LiDAR) data. For Part B, we use a reversed solar ray casting approach we call the ‘Subtracto-Sun’ which allows a user to specify a time range (e.g. 9 a.m.–4 p.m.) for multiple days of the year and subsequent generation of 3D maximum building height development envelopes. The output of this system was used by the local government to inform a proposed planning policy amendment for the City of Melbourne. The findings illustrate the potential for urban professionals to use the system to rapidly assess shadow impacts for existing and proposed, contextually accurate, large, complex urban environments with high levels of geometric and temporal details. The presented results are significant in that we develop and apply our spatio-temporal decision-support approach to a local government area, successfully informing planning height restriction decisions to protect daylight amenity of public open spaces in need of protection. Our method for setting development height restrictions allows for higher density to be achieved, while not increasing the overshadowing of critical open space infrastructure during designated times.

Laserfarm – A high-throughput workflow for generating geospatial data products of ecosystem structure from airborne laser scanning point clouds

Quantifying ecosystem structure is of key importance for ecology, conservation, restoration, and biodiversity monitoring because the diversity, geographic distribution and abundance of animals, plants and other organisms is tightly linked to the physical structure of vegetation and associated microclimates. Light Detection And Ranging (LiDAR) — an active remote sensing technique — can provide detailed and high resolution information on ecosystem structure because the laser pulse emitted from the sensor and its subsequent return signal from the vegetation (leaves, branches, stems) delivers three-dimensional point clouds from which metrics of vegetation structure (e.g. ecosystem height, cover, and structural complexity) can be derived. However, processing 3D LiDAR point clouds into geospatial data products of ecosystem structure remains challenging across broad spatial extents due to the large volume of national or regional point cloud datasets (typically multiple terabytes consisting of hundreds of billions of points). Here, we present a high-throughput workflow called ‘Laserfarm’ enabling the efficient, scalable and distributed processing of multi-terabyte LiDAR point clouds from national and regional airborne laser scanning (ALS) surveys into geospatial data products of ecosystem structure. Laserfarm is a free and open-source, end-to-end workflow which contains modular pipelines for the re-tiling, normalization, feature extraction and rasterization of point cloud information from ALS and other LiDAR surveys. The workflow is designed with horizontal scalability and can be deployed with distributed computing on different infrastructures, e.g. a cluster of virtual machines. We demonstrate the Laserfarm workflow by processing a country-wide multi-terabyte ALS dataset of the Netherlands (covering ∼34,000 km2 with ∼700 billion points and ∼ 16 TB uncompressed LiDAR point clouds) into 25 raster layers at 10 m resolution capturing ecosystem height, cover and structural complexity at a national extent. The Laserfarm workflow, implemented in Python and available as Jupyter Notebooks, is applicable to other LiDAR datasets and enables users to execute automated pipelines for generating consistent and reproducible geospatial data products of ecosystems structure from massive amounts of LiDAR point clouds on distributed computing infrastructures, including cloud computing environments. We provide information on workflow performance (including total CPU times, total wall-time estimates and average CPU times for single files and LiDAR metrics) and discuss how the Laserfarm workflow can be scaled to other LiDAR datasets and computing environments, including remote cloud infrastructures. The Laserfarm workflow allows a broad user community to process massive amounts of LiDAR point clouds for mapping vegetation structure, e.g. for applications in ecology, biodiversity monitoring and ecosystem restoration.

RETRACTED: SHREC 2021: 3D point cloud change detection for street scenes

The rapid development of 3D acquisition devices enables us to collect billions of points in a few hours. However, the analysis of the output data is a challenging task, especially in the field of 3D point cloud change detection. In this Shape Retrieval Challenge (SHREC) track, we provide a street-scene dataset for 3D point cloud change detection. The dataset consists of 866 3D object pairs in year 2016 and 2020 from 78 large-scale street scene 3D point clouds. Our goal is to detect the changes from multi-temporal point clouds in a complex street environment. We compare three methods on this benchmark, with one handcrafted (PoChaDeHH) and the other two learning-based (HGI-CD and SiamGCN). The results show that the handcrafted algorithm has balanced performance over all classes, while learning-based methods achieve overwhelming performance but suffer from the class-imbalanced problem and may fail on minority classes. The randomized oversampling metric applied in SiamGCN can alleviate this problem. Also, different siamese network architecture in HGI-CD and SiamGCN contribute to the designing of a network for the 3D change detection task.

A Local Graph-Based Structure for Processing Gigantic Aggregated 3D Point Clouds

We present an original workflow for structuring a point cloud generated from several scans. Our representation is based on a set of local graphs. Each graph is constructed from the depth map provided by each scan. The graphs are then connected together via the overlapping areas, and careful consideration of the redundant points in these regions leads to a piecewise and globally consistent structure for the underlying surface sampled by the point cloud. The proposed workflow allows structuring aggregated point clouds, scan after scan, whatever the number of acquisitions and the number of points per acquisition, even on computers with very limited memory capacities. To show that our structure can be highly relevant for the community, where the gigantic amount of data represents a real scientific challenge per se, we present an algorithm based on this structure capable of resampling billions of points on standard computers. This application is particularly attractive for simplifying and visualizing gigantic point clouds representing very large-scale scenes (buildings, urban scenes, historical sites...), which often require a prohibitive number of points to describe them accurately.

EFFICIENT LOADING AND VISUALIZATION OF MASSIVE FEATURE-RICH POINT CLOUDS WITHOUT HIERARCHICAL ACCELERATION STRUCTURES

Abstract. Nowadays, point clouds are the standard product when capturing reality independent of scale and measurement technique. Especially, Dense Image Matching (DIM) and Laser Scanning (LS) are state of the art capturing methods for a great variety of applications producing detailed point clouds up to billions of points. In-depth analysis of such huge point clouds typically requires sophisticated spatial indexing structures to support potentially long-lasting automated non-interactive processing tasks like feature extraction, semantic labelling, surface generation, and the like. Nevertheless, a visual inspection of the point data is often necessary to obtain an impression of the scene, roughly check for completeness, quality, and outlier rates of the captured data in advance. Also intermediate processing results, containing additional per-point computed attributes, may require visual analyses to draw conclusions or to parameterize further processing. Over the last decades a variety of commercial, free, and open source viewers have been developed that can visualise huge point clouds and colorize them based on available attributes. However, they have either a poor loading and navigation performance, visualize only a subset of the points, or require the creation of spatial indexing structures in advance. In this paper, we evaluate a progressive method that is capable of rendering any point cloud that fits in GPU memory in real time without the need of time consuming hierarchical acceleration structure generation. In combination with our multi-threaded LAS and LAZ loaders, we achieve load performance of up to 20 million points per second, display points already while loading, support flexible switching between different attributes, and rendering up to one billion points with visually appealing navigation behaviour. Furthermore, loading times of different data sets for different open source and commercial software packages are analysed.

A charge density prediction model for hydrocarbons using deep neural networks

The electronic charge density distribution ρ(r) of a given material is among the most fundamental quantities in quantum simulations from which many large scale properties and observables can be calculated. Conventionally, ρ(r) is obtained using Kohn–Sham density functional theory (KS-DFT) based methods. But, the high computational cost of KS-DFT renders it intractable for systems involving thousands/millions of atoms. Thus, recently there has been efforts to bypass expensive KS equations, and directly predict ρ(r) using machine learning (ML) based methods. Here, we build upon one such scheme to create a robust and reliable ρ(r) prediction model for a diverse set of hydrocarbons, involving huge chemical and morphological complexity /(saturated, unsaturated molecules, cyclo-groups and amorphous and semi-crystalline polymers). We utilize a grid-based fingerprint to capture the atomic neighborhood around an arbitrary point in space, and map it to the reference ρ(r) obtained from standard DFT calculations at that point. Owing to the grid-based learning, dataset sizes exceed billions of points, which is trained using deep neural networks in conjunction with a incremental learning based approach. The accuracy and transferability of the ML approach is demonstrated on not only a diverse test set, but also on a completely unseen system of polystyrene under different strains. Finally, we note that the general approach adopted here could be easily extended to other material systems, and can be used for quick and accurate determination of ρ(r) for DFT charge density initialization, computing dipole or quadrupole, and other observables for which reliable density functional are known.

Empirical Bayesian kriging implementation and usage

We described the key features of the pragmatic geostatistical methodology aiming at resolving the following drawbacks of classical geostatistical models: assuming that the data is the realization of a stationary process; assuming that the data values are distributed according to Gaussian distribution; describing the data with a single generating model; not accounting for the model uncertainty in prediction; and not supporting coincident data and individual measurement errors. Our variant of empirical Bayesian kriging (EBK) is a fast and reliable solution for both automatic and interactive data interpolation. It can be used for interpolation of very large datasets up to billions of points. The following features are discussed: the informative prior distribution construction and usage; automatic data transformation of the dependent variable into a Gaussian distribution; data subsetting and merging the estimated models; and interpolation over large areas on the earth's surface. We conducted one simulation experiment and two case studies using highly variable data to investigate the EBK predicting quality.

Concepts and techniques for web-based visualization and processing of massive 3D point clouds with semantics

3D point cloud technology facilitates the automated and highly detailed acquisition of real-world environments such as assets, sites, and countries. We present a web-based system for the interactive exploration and inspection of arbitrary large 3D point clouds. Our approach is able to render 3D point clouds with billions of points using spatial data structures and level-of-detail representations. Point-based rendering techniques and post-processing effects are provided to enable task-specific and data-specific filtering, e.g., based on semantics. A set of interaction techniques allows users to collaboratively work with the data (e.g., measuring distances and annotating). Additional value is provided by the system’s ability to display additional, context-providing geodata alongside 3D point clouds and to integrate processing and analysis operations. We have evaluated the presented techniques and in case studies and with different data sets from aerial, mobile, and terrestrial acquisition with up to 120 billion points to show their practicality and feasibility.

HiSpatialCluster: A novel high‐performance software tool for clustering massive spatial points

AbstractIn the era of big data, spatial clustering is a very important means for geo‐data analysis. When clustering big geo‐data such as social media check‐in data, geotagged photos, and taxi trajectory points, traditional spatial clustering algorithms are facing more challenges. On the one hand, existing spatial clustering tools cannot support the clustering of massive point sets; on the other hand, there is no perfect solution for self‐adaptive spatial clustering. In order to achieve clustering of millions or even billions of points adaptively, a new spatial clustering tool—HiSpatialCluster—was proposed, in which the CFSFDP (clustering by fast search and finding density peaks) idea to find cluster centers and the DBSCAN (density‐based spatial clustering of applications with noise) idea of density‐connect filtering for classification are introduced. The tool’s source codes and other resources have been released on Github, and experimental evaluation was performed through clustering massive taxi trajectory points and Flickr geotagged photos in Beijing, China. The spatial clustering results were compared with those through K‐means and DBSCAN as well. As a spatial clustering tool, HiSpatialCluster is expected to play a fundamental role in big geo‐data research. First, this tool enables clustering adaptively on massive point datasets with uneven spatial density distribution. Second, the density‐connect filter method is applied to generate homogeneous analysis units from geotagged data. Third, the tool is accelerated by both parallel CPU and GPU computing so that millions or even billions of points can be clustered efficiently.

Model of Point Cloud Data Management System in Big Data Paradigm

Modern geoinformation technologies for collecting and processing data, such as laser scanning or photogrammetry, can generate point clouds with billions of points. They provide abundant information that can be used for different types of analysis. Due to its characteristics, the point cloud is often viewed as a special type of geospatial data. In order to efficiently manage such volumes of data, techniques based on a computer cluster have to be used. The Apache Spark framework has proven to be a solution for efficient processing of large volumes of data. This paper thoroughly examines the representation of point cloud data type using Apache Spark constructs. The common operations over point clouds, range queries and k-nearest neighbors queries (kNN) are implemented using Apache Spark DataFrame Application Programming Interface (API). It enabled the design of point cloud related user defined types (UDT) and user defined functions (UDF). The structure of the point cloud for efficient storing in Big Data key-value stores was analyzed and described. The methods presented in this paper were compared to PostgreSQL RDBMS, and the results were discussed.

Remote visual analysis of large turbulence databases at multiple scales

The remote analysis and visualization of raw large turbulence datasets is challenging. Current accurate direct numerical simulations (DNS) of turbulent flows generate datasets with billions of points per time-step and several thousand time-steps per simulation. Until recently, the analysis and visualization of such datasets was restricted to scientists with access to large supercomputers. The public Johns Hopkins Turbulence database simplifies access to multi-terabyte turbulence datasets and facilitates the computation of statistics and extraction of features through the use of commodity hardware. We present a framework designed around wavelet-based compression for high-speed visualization of large datasets and methods supporting multi-resolution analysis of turbulence. By integrating common technologies, this framework enables remote access to tools available on supercomputers and over 230 terabytes of DNS data over the Web. The database toolset is expanded by providing access to exploratory data analysis tools, such as wavelet decomposition capabilities and coherent feature extraction.

On the Effect of Geometries Simplification on Geo-spatial Link Discovery

Link discovery is central to the integration and use of data across RDF knowledge bases. Geospatial information is increasingly represented according to the Linked Data principles. Resources within such datasets are described by means of vector geometry, where link discovery approaches have to deal with millions of point sets consisting of billions of points. In this paper, we study the effect of simplifying the resources’ geometries on runtime and F-measure of link discovery approaches. In particular, we evaluate link discovery approaches for computing the point-set distances as well as the topological relations among RDF resources with geospatial representation. The results obtained on two different real datasets suggest that most geospatial link discovery approaches achieve up to 67x speedup using simplification, while the average loss in their F-measure is less than 15%. Our implementation is open-source and available at http://github.com/dice-group/limes.

CLASSIFICATION OF MOBILE LASER SCANNING POINT CLOUDS FROM HEIGHT FEATURES

Abstract. The demand for 3D maps of cities and road networks is steadily growing and mobile laser scanning (MLS) systems are often the preferred geo-data acquisition method for capturing such scenes. Because MLS systems are mounted on cars or vans they can acquire billions of points of road scenes within a few hours of survey. Manual processing of point clouds is labour intensive and thus time consuming and expensive. Hence, the need for rapid and automated methods for 3D mapping of dense point clouds is growing exponentially. The last five years the research on automated 3D mapping of MLS data has tremendously intensified. In this paper, we present our work on automated classification of MLS point clouds. In the present stage of the research we exploited three features – two height components and one reflectance value, and achieved an overall accuracy of 73 %, which is really encouraging for further refining our approach.

DETECTION OF GEOMETRIC KEYPOINTS AND ITS APPLICATION TO POINT CLOUD COARSE REGISTRATION

Acquisition of large scale scenes, frequently, involves the storage of large amount of data, and also, the placement of several scan positions to obtain a complete object. This leads to a situation with a different coordinate system in each scan position. Thus, a preprocessing of it to obtain a common reference frame is usually needed before analysing it. Automatic point cloud registration without locating artificial markers is a challenging field of study. The registration of millions or billions of points is a demanding task. Subsampling the original data usually solves the situation, at the cost of reducing the precision of the final registration. In this work, a study of the subsampling via the detection of keypoints and its capability to apply in coarse alignment is performed. The keypoints obtained are based on geometric features of each individual point, and are extracted using the Difference of Gaussians approach over 3D data. The descriptors include features as eigenentropy, change of curvature and planarity. Experiments demonstrate that the coarse alignment, obtained through these keypoints outperforms the coarse registration root mean squared error of an operator by 3 - 5 cm. The applicability of these keypoints is tested and verified in five different case studies.

DETECTION OF GEOMETRIC KEYPOINTS AND ITS APPLICATION TO POINT CLOUD COARSE REGISTRATION

Abstract. Acquisition of large scale scenes, frequently, involves the storage of large amount of data, and also, the placement of several scan positions to obtain a complete object. This leads to a situation with a different coordinate system in each scan position. Thus, a preprocessing of it to obtain a common reference frame is usually needed before analysing it. Automatic point cloud registration without locating artificial markers is a challenging field of study. The registration of millions or billions of points is a demanding task. Subsampling the original data usually solves the situation, at the cost of reducing the precision of the final registration. In this work, a study of the subsampling via the detection of keypoints and its capability to apply in coarse alignment is performed. The keypoints obtained are based on geometric features of each individual point, and are extracted using the Difference of Gaussians approach over 3D data. The descriptors include features as eigenentropy, change of curvature and planarity. Experiments demonstrate that the coarse alignment, obtained through these keypoints outperforms the coarse registration root mean squared error of an operator by 3 - 5 cm. The applicability of these keypoints is tested and verified in five different case studies.

SIDELOADING – INGESTION OF LARGE POINT CLOUDS INTO THE APACHE SPARK BIG DATA ENGINE

Abstract. In the geospatial domain we have now reached the point where data volumes we handle have clearly grown beyond the capacity of most desktop computers. This is particularly true in the area of point cloud processing. It is therefore naturally lucrative to explore established big data frameworks for big geospatial data. The very first hurdle is the import of geospatial data into big data frameworks, commonly referred to as data ingestion. Geospatial data is typically encoded in specialised binary file formats, which are not naturally supported by the existing big data frameworks. Instead such file formats are supported by software libraries that are restricted to single CPU execution. We present an approach that allows the use of existing point cloud file format libraries on the Apache Spark big data framework. We demonstrate the ingestion of large volumes of point cloud data into a compute cluster. The approach uses a map function to distribute the data ingestion across the nodes of a cluster. We test the capabilities of the proposed method to load billions of points into a commodity hardware compute cluster and we discuss the implications on scalability and performance. The performance is benchmarked against an existing native Apache Spark data import implementation.