Scale Point Cloud Research Articles

Novel Class Discovery for Representation of Real-World Heritage Data as Neural Radiance Fields (Student Abstract)

Neural Radiance Fields (NeRF) have been extensively explored as a leading approach for modeling and representing 3D data across various domains. Their ability to capture arbitrary scale point clouds and generate novel views makes them particularly valuable for digitizing cultural heritage sites. However, despite their impressive rendering capabilities, prior methods have often overlooked a significant real-world challenge: handling open-world scenarios characterized by unstructured data containing multiple classes in a single set of unlabeled images. To address this challenge, we propose a novel method NCD-NeRF that leverages Novel-Class Discovery to effectively tackle the complexities inherent in real-world data with unlabeled classes while excelling in producing high-quality NeRF representation. To validate our approach, we conducted a benchmarking analysis using a custom-collected dataset featuring UNESCO World Heritage sites in India. We observe that our proposed NCD-NeRF can parallely discover novel classes and render high-quality 3D volumes.

A METHODOLOGY FOR CAVE FLOOR BASEMAP SYNTHESIS FROM POINT CLOUD DATA: A CASE STUDY OF SLAM-BASED LIDAR AT LAS CUEVAS, BELIZE

Abstract. Creating cave maps is an essential part of cave research. Traditional cartographic efforts are extremely time consuming and subjective, motivating the development of new techniques using terrestrial lidar scanners and mobile lidar systems. However, processing the large point clouds from these scanners to produce detailed, yet manageable “maps” remains a challenge. In this work, we present a methodology for synthesizing a basemap representing the cave floor from large scale point clouds, based on a case study of a SLAM-based lidar data acquisition from a cave system in the archaeological site of Las Cuevas, Belize. In 4 days of fieldwork, the 335 m length of the cave system was scanned, resulting in a point cloud of 4.1 billion points, with 1.6 billion points classified as part of the cave floor. This point cloud was processed to produce a basemap that can be used in GIS, where natural and anthropogenic features are clearly visible and can be traced to create accurate 2D maps similar to traditional cartography.

Light bottle transformer based large scale point cloud classification

Light bottle transformer based large scale point cloud classification

Fast color point cloud registration based on virtual viewpoint image

With the increase of point cloud scale, the time required by traditional ICP-related point cloud registration methods increases dramatically, which cannot meet the registration requirements of large-scale point clouds. In this paper, a fast registration technique for large scale point clouds based on virtual viewpoint image generation is studied. Firstly, the projection image of color point cloud is generated by virtual viewpoint. Then, the feature is extracted based on ORB and the rotation and translation matrix is calculated. The experimental results show that the registration time of the proposed method is about 1s when the size of the point cloud is from 300,000 to 2 million, which is improved by 17–258 times compared with the traditional ICP registration method, and the registration error is reduced by 80% from ICP 5.0 to 1.0. This paper provides a new idea and method for large-scale color point cloud registration.

Adaptive feature-conserving compression for large scale point clouds

In this work, we introduce a practical method for reducing big point clouds of buildings and infrastructure. The proposed method introduces bilateral filtering with a tailored set of evaluation functions to conserve maximum information. The statistical parameters necessary for our model are selected by examining various point properties of a comprehensive dataset. The dataset contains artificial, photogrammetric and laser-scanned point clouds and has been made publicly available. For verification, we showcase our filtering method by preserving more information than voxel grid or density filters, enabling even sparser photogrammetric datasets. Finally, we discuss some encoding strategies as well as the best balance between size and resolution.

EDC-Net: Edge Detection Capsule Network for 3D Point Clouds

Edge features in point clouds are prominent due to the capability of describing an abstract shape of a set of points. Point clouds obtained by 3D scanner devices are often immense in terms of size. Edges are essential features in large scale point clouds since they are capable of describing the shapes in down-sampled point clouds while maintaining the principal information. In this paper, we tackle challenges of edge detection tasks in 3D point clouds. To this end, we propose a novel technique to detect edges of point clouds based on a capsule network architecture. In this approach, we define the edge detection task of point clouds as a semantic segmentation problem. We built a classifier through the capsules to predict edge and non-edge points in 3D point clouds. We applied a weakly-supervised learning approach in order to improve the performance of our proposed method and built in the capability of testing the technique in wider range of shapes. We provide several quantitative and qualitative experimental results to demonstrate the robustness of our proposed EDC-Net for edge detection in 3D point clouds. We performed a statistical analysis over the ABC and ShapeNet datasets. Our numerical results demonstrate the robust and efficient performance of EDC-Net.

Unsupervised semantic and instance segmentation of forest point clouds

Terrestrial Laser Scanning (TLS) has been increasingly used in forestry applications including forest inventory and plant ecology. Tree biophysical properties such as leaf area distributions and wood volumes can be accurately estimated from TLS point clouds. In these applications, a prerequisite is to properly understand the information content of large scale point clouds (i.e., semantic labelling of point clouds), so that tree-scale attributes can be retrieved. Currently, this requirement is undergoing laborious and time consuming manual works. In this work, we jointly address the problems of semantic and instance segmentation of forest point clouds. Specifically, we propose an unsupervised pipeline based on a structure called superpoint graph, to simultaneously perform two tasks: single tree isolation and leaf-wood classification. The proposed method is free from restricted assumptions of forest types. Validation using simulated data resulted in a mean Intersection over Union (mIoU) of 0.81 for single tree isolation, and an overall accuracy of 87.7% for leaf-wood classification. The single tree isolation led to a relative root mean square error (RMSE%) of 2.9% and 19.8% for tree height and crown diameter estimations, respectively. Comparisons with existing methods on other benchmark datasets showed state-of-the-art results of our method on both single tree isolation and leaf-wood classification tasks. We provide the entire framework as an open-source tool with an end-user interface. This study closes the gap for using TLS point clouds to quantify tree-scale properties in large areas, where automatic interpretation of the information content of TLS point clouds remains a crucial challenge.

Large-scale point cloud contour extraction via 3D guided multi-conditional generative adversarial network

As one of the most important features for human perception, contours are widely used in many graphics and mapping applications. However, for large outdoor scale point clouds, contour extraction is considerably challenging due to the huge, unstructured and irregular point space, thus leading to massive failure for existing approaches. In this paper, to generate contours consistent with human perception for outdoor scenes, we propose, for the first time, 3D guided multi-conditional GAN (3D-GMcGAN), a deep neural network based contour extraction network for large scale point clouds. Specifically, two ideas are proposed to enable the network to learn the distributions of labeled samples. First, a parametric space based framework is proposed via a novel similarity measurement of two parametric models. Such a framework significantly compresses the huge point data space, thus making it much easier for the network to “remember” target distribution. Second, to prevent network loss in the huge solution space, a guided learning framework is designed to assist finding the target contour distribution via an initial guidance. To evaluate the effectiveness of the pro-posed network, we open-sourced the first, to our knowledge, dataset for large scale point cloud with contour annotation information. Experimental results demonstrate that 3D-GMcGAN efficiently generates contours for the data with more than ten million points (about several minutes), while avoiding ad hoc stages or parameters. Also, the proposed framework produces minimal outliers and pseudo-contours, as suggested by comparisons with the state-of-the-art approaches.

Fast Cylinder Shape Matching Using Random Sample Consensus in Large Scale Point Cloud

In this paper, an algorithm is proposed that can perform cylinder type matching faster than the existing method in point clouds that represent space. The existing matching method uses Hough transform and completes the matching through preprocessing such as noise filtering, normal estimation, and segmentation. The proposed method completes the matching through the methodology of random sample consensus (RANSAC) and principal component analysis (PCA). Cylindrical pipe estimation is based on two mathematical models that compute the parameters and combine the results to predict spheres and lines. RANSAC fitting computes the center and radius of the sphere, which can be the radius of the cylinder axis and finds straight and curved areas through PCA. This allows fast matching without normal estimation and segmentation. Linear and curved regions are distinguished by a discriminant using eigenvalues. The linear region is the sum of the vectors of linear candidates, and the curved region is matched by a Catmull–Rom spline. The proposed method is expected to improve the work efficiency of the reverse design by matching linear and curved cylinder estimation without vertical/horizontal constraint and segmentation. It is also more than 10 times faster while maintaining the accuracy of the conventional method.

高维正交子空间映射的尺度点云配准算法

高维正交子空间映射的尺度点云配准算法

Advances in Automated Volume Analysis Based on Large Scale Pointclouds

Advances in Automated Volume Analysis Based on Large Scale Pointclouds

AIRBORNE LIGHT DETECTION AND RANGING (LIDAR) DERIVED DEFORMATION FROM THE MW 6.0 24 AUGUST, 2014 SOUTH NAPA EARTHQUAKE ESTIMATED BY TWO AND THREE DIMENSIONAL POINT CLOUD CHANGE DETECTION TECHNIQUES

Abstract. Remote sensing via LiDAR (Light Detection And Ranging) has proven extremely useful in both Earth science and hazard related studies. Surveys taken before and after an earthquake for example, can provide decimeter-level, 3D near-field estimates of land deformation that offer better spatial coverage of the near field rupture zone than other geodetic methods (e.g., InSAR, GNSS, or alignment array). In this study, we compare and contrast estimates of deformation obtained from different pre and post-event airborne laser scanning (ALS) data sets of the 2014 South Napa Earthquake using two change detection algorithms, Iterative Control Point (ICP) and Particle Image Velocimetry (PIV). The ICP algorithm is a closest point based registration algorithm that can iteratively acquire three dimensional deformations from airborne LiDAR data sets. By employing a newly proposed partition scheme, “moving window,” to handle the large spatial scale point cloud over the earthquake rupture area, the ICP process applies a rigid registration of data sets within an overlapped window to enhance the change detection results of the local, spatially varying surface deformation near-fault. The other algorithm, PIV, is a well-established, two dimensional image co-registration and correlation technique developed in fluid mechanics research and later applied to geotechnical studies. Adapted here for an earthquake with little vertical movement, the 3D point cloud is interpolated into a 2D DTM image and horizontal deformation is determined by assessing the cross-correlation of interrogation areas within the images to find the most likely deformation between two areas. Both the PIV process and the ICP algorithm are further benefited by a presented, novel use of urban geodetic markers. Analogous to the persistent scatterer technique employed with differential radar observations, this new LiDAR application exploits a classified point cloud dataset to assist the change detection algorithms. Ground deformation results and statistics from these techniques are presented and discussed here with supplementary analyses of the differences between techniques and the effects of temporal spacing between LiDAR datasets. Results show that both change detection methods provide consistent near field deformation comparable to field observed offsets. The deformation can vary in quality but estimated standard deviations are always below thirty one centimeters. This variation in quality differentiates the methods and proves that factors such as geodetic markers and temporal spacing play major roles in the outcomes of ALS change detection surveys.

AIRBORNE LIGHT DETECTION AND RANGING (LIDAR) DERIVED DEFORMATION FROM THE MW 6.0 24 AUGUST, 2014 SOUTH NAPA EARTHQUAKE ESTIMATED BY TWO AND THREE DIMENSIONAL POINT CLOUD CHANGE DETECTION TECHNIQUES

Remote sensing via LiDAR (Light Detection And Ranging) has proven extremely useful in both Earth science and hazard related studies. Surveys taken before and after an earthquake for example, can provide decimeter-level, 3D near-field estimates of land deformation that offer better spatial coverage of the near field rupture zone than other geodetic methods (e.g., InSAR, GNSS, or alignment array). In this study, we compare and contrast estimates of deformation obtained from different pre and post-event airborne laser scanning (ALS) data sets of the 2014 South Napa Earthquake using two change detection algorithms, Iterative Control Point (ICP) and Particle Image Velocimetry (PIV). The ICP algorithm is a closest point based registration algorithm that can iteratively acquire three dimensional deformations from airborne LiDAR data sets. By employing a newly proposed partition scheme, “moving window,” to handle the large spatial scale point cloud over the earthquake rupture area, the ICP process applies a rigid registration of data sets within an overlapped window to enhance the change detection results of the local, spatially varying surface deformation near-fault. The other algorithm, PIV, is a well-established, two dimensional image co-registration and correlation technique developed in fluid mechanics research and later applied to geotechnical studies. Adapted here for an earthquake with little vertical movement, the 3D point cloud is interpolated into a 2D DTM image and horizontal deformation is determined by assessing the cross-correlation of interrogation areas within the images to find the most likely deformation between two areas. Both the PIV process and the ICP algorithm are further benefited by a presented, novel use of urban geodetic markers. Analogous to the persistent scatterer technique employed with differential radar observations, this new LiDAR application exploits a classified point cloud dataset to assist the change detection algorithms. Ground deformation results and statistics from these techniques are presented and discussed here with supplementary analyses of the differences between techniques and the effects of temporal spacing between LiDAR datasets. Results show that both change detection methods provide consistent near field deformation comparable to field observed offsets. The deformation can vary in quality but estimated standard deviations are always below thirty one centimeters. This variation in quality differentiates the methods and proves that factors such as geodetic markers and temporal spacing play major roles in the outcomes of ALS change detection surveys.

Interactive Pickup of Three-dimensional Fine Point Cloud Based on GPU

Three-dimensional fine point cloud has gradually become a key data source of three-dimensional model. The large scale point cloud interactive quick pick up is a kind of important operation in the point cloud data processing and applications. Since the point cloud model is composed of massive points, the speed of ordinary picking method is limited. A GPU-based point cloud picking algorithm was thus presented to solve the problem. The basic idea of the algorithm is that by spatial transformation converting the point cloud to screen space, and then, the point was calculated which is the nearest to the mouse click point in screen space. The GPU's parallel computing capabilities were used to achieve spatial transformation and distance comparison by compute shader in this algorithm. So the speed of the pickup has been increased. The results show that compared with the CPU, the pickup method based on GPU has greater speed advantage. Especially for the point cloud over 4 million points, the speed of the pickup has been increased 2-3 times faster.

基于运动恢复的双目视觉三维重建系统设计

An 3D reconstruction system based on binocular stereo vision was improved to obtain higher accuracy and larger scale of 3D reconstruction. A disparity refinement procedure was introduced to the system, so that the matching cost between the original parallax and the adjacent disparity was fitted into a quadratic curve and to re-find a more accurate parallax for the curve. Then, the motion recovery calculation was further applies to estimation of the camera motion matrices and the tracking points and camera motion matrices were taken as the parameters to construct the energy function. Furthermore, the energy function was optimized to reduce errors effectively and reconstruct the motion matrix accurately. Experimental results indicate that the proposed refinement procedure effectively improves the accuracy of reconstruction of point clouds and reduces the error by 16.3 percent in average with avoiding the scale point cloud phenomenon. As the motion recovery calculation steps recover the camera motion matrix accurately, the optimized method reduces the mean re-projection error by 95.5 percent. The point clouds achieved from different angle images are no longer isolated and the reconstructed models are integrated naturally.

A DATA MANAGEMENT METHOD FOR EFFICIENT SEARCH AND RENDERING OF MULTIPLE LARGE SCALE POINT CLOUDS

Abstract. Recently, with the progress of laser scanning technology, it has become possible to scan several environments such as buildings, roads, and cities easily and at a low cost. There are typically three scanning methods: TLS (Terrestrial Laser Scanning), MMS (Mobile Mapping System), and ALS (Airborne Laser Scanning), and the environments are scanned for measurement, modelling, and simulation etc. Therefore, an efficient and comprehensive data management method, which can handle these point clouds in various applications is strongly required. In this paper, we describe a method of data management for efficient search and rendering of multiple large scale point clouds. First, we define a general point cloud file format including the octree and the quantization structure. Next, the data management method enabling an efficient search and rendering of the point clouds is proposed. Finally, the effectiveness of our method is verified using several actual data sets.

Computation of approximate geodesics on point cloud

In order to compute approximate geodesic efficiently between two points on point cloud,a weighted graph was constructed by splitting point cloud along the Cartesian coordinate axes,thus initial approximate geodesic between any two given points could be computed out using Dijkstra's algorithm.Then the conjugate gradient method was adopted to minimize the energy function defined;finally,approximate geodesic could be obtained after some iterative steps.This proposed algorithm avoids meshing or reconstructing the point cloud to be local or global surface,and it is suitable for computing geodesic on large scale point cloud.