Surface Reconstruction from Point Clouds by Learning Predictive Context Priors

This paper presents an integrated framework for surface reconstruction capable of handling large scale clouds of points. This framework is based on two proposed methods for implicit surface fitting and polygonization to convert a cloud of unorganized points into an optimized surface. The proposed fitting method employs the partition of unity (POU) method associated with the radial basis functions (RBF) over a distributed computing environment to facilitate and speedup fitting of large scale clouds without any data reduction to preserve all the surface details. Moreover, an innovative adaptive mesh refinement (AMR) based method is proposed for implicit surface polygonization. This method steers adaptive volume sampling via a series of optimization criteria to provide accurate and optimized surfaces with minimum number of polygons. The experimental results for the considered test models showed an average reduction of 60% in fitting time using 16 processing nodes and 90% in polygonization time on the master node only against other traditional methods with better performance.

Surface reconstruction from 3D point clouds has a wide range of applications. In this paper, we focus on the reconstruction from raw, sparse point clouds. Although some existing methods work on this topic, the results often suffer from geometric defects. To solve this problem, we propose a novel method that optimizes a neural network (referred to as signed distance function) to fit the Signed Distance Field (SDF) from sparse point clouds. The signed distance function is optimized by projecting query points to its iso-surface accordingly. Our key idea is to encourage both the direction and distance of projection to be correct through the supervision provided by a nearest neighbor prior. In addition, we mitigate the error propagated from the prior function by augmenting the low-frequency components in the input. In our implementation, the nearest neighbor prior is trained with a large-scale local geometry dataset, and the positional encoding with a specified spectrum is used as a regularization for the optimization process. Experiments on the ShapeNetCore dataset demonstrate that our method achieves better accuracy than SDF-based methods while preserving smoothness.

In reverse engineering, the point cloud data processing is the key, it will directly affect the quality and efficiency of the subsequent reconstruction of the surface. However, the amount of data obtained by three-dimensional laser scanning is often larger and redundant, which will increase the difficulty of the surface reconstruction, therefore, the point cloud data is needed to streamline. Depth analysis of the efficiency and accuracy of curvature grid method, a secondary streamlined algorithm is proposed, which take use of non-uniform grid streamline law based on octree to streamline to the point cloud, and re-use curvature streamline method based on B-spline surface fitting to streamline it. This method, whose streamline ratio is relatively high, can effectively retain the feature point cloud data, providing a great convenience for the follow-up of reverse engineering.

Surface reconstruction is a challenging task when input point clouds, especially real scans, are noisy and lack normals. Observing that the Multilayer Perceptron (MLP) and the implicit moving least-square function (IMLS) provide a dual representation of the underlying surface, we introduce Neural-IMLS, a novel approach that directly learns a noise-resistant signed distance function (SDF) from unoriented raw point clouds in a self-supervised manner. In particular, IMLS regularizes MLP by providing estimated SDFs near the surface and helps enhance its ability to represent geometric details and sharp features, while MLP regularizes IMLS by providing estimated normals. We prove that at convergence, our neural network produces a faithful SDF whose zero-level set approximates the underlying surface due to the mutual learning mechanism between the MLP and the IMLS. Extensive experiments on various benchmarks, including synthetic and real scans, show that Neural-IMLS can reconstruct faithful shapes even with noise and missing parts. The source code can be found at https://github.com/bearprin/Neural-IMLS.

The fusion of light detection and ranging (LiDAR) and camera provides comprehensive information about the depth and color texture of objects, finding wide applications in autonomous driving, three-dimensional (3D) scene reconstruction, and other fields. However, point clouds acquired via LiDAR are susceptible to noise, impacting downstream tasks like surface reconstruction and object recognition demanding high-quality point cloud outputs from 3D imaging systems. This paper proposes a novel 3D fusion imaging technique that enhances the quality of the fused point clouds from three distinct aspects, improving the accuracy and precision of point clouds by employing matched filter method, designing a preprocessing workflow for point clouds and images to enable their effective fusion, and smoothing noise and accentuating texture features using a filter algorithm on fused information. The method of matched filtering has been employed to reconstruct object distance information from photon count-time histograms, resulting in an improvement of approximately 120% compared to the method of directly calculating the average photon flight time. The point cloud filter proposed utilizing fused information demonstrates superior denoising capabilities compared to the state-of-the-art and has the capability to cluster points with similar features to highlight the texture characteristics of the point cloud.

The existing surface reconstruction algorithms currently reconstruct large amounts of mesh data. Consequently, many of these algorithms cannot meet the efficiency requirements of real-time data transmission in a web environment. This paper proposes a lightweight surface reconstruction method for online 3D scanned point cloud data oriented toward 3D printing. The proposed online lightweight surface reconstruction algorithm is composed of a point cloud update algorithm (PCU), a rapid iterative closest point algorithm (RICP), and an improved Poisson surface reconstruction algorithm (IPSR). The generated lightweight point cloud data are pretreated using an updating and rapid registration method. The Poisson surface reconstruction is also accomplished by a pretreatment to recompute the point cloud normal vectors; this approach is based on a least squares method, and the postprocessing of the PDE patch generation was based on biharmonic-like fourth-order PDEs, which effectively reduces the amount of reconstructed mesh data and improves the efficiency of the algorithm. This method was verified using an online personalized customization system that was developed with WebGL and oriented toward 3D printing. The experimental results indicate that this method can generate a lightweight 3D scanning mesh rapidly and efficiently in a web environment.

3D object detection from raw and sparse point clouds has been far less treated to date, compared with its 2D counterpart. In this paper, we propose a novel framework called FVNet for 3D front-view proposal generation and object detection from point clouds. It consists of two stages: generation of front-view proposals and estimation of 3D bounding box parameters. Instead of generating proposals from camera images or bird's-eye-view maps, we first project point clouds onto a cylindrical surface to generate front-view feature maps which retains rich information. We then introduce a proposal generation network to predict 3D region proposals from the generated maps and further extrude objects of interest from the whole point cloud. Finally, we present another network to extract the point-wise features from the extruded object points and regress the final 3D bounding box parameters in the canonical coordinates. Our framework achieves real-time performance with 12ms per point cloud sample. Extensive experiments on the 3D detection benchmark KITTI show that the proposed architecture outperforms state-of-the-art techniques which take either camera images or point clouds as input, in terms of accuracy and inference time.

SAANet: Spatial adaptive alignment network for object detection in automatic driving

With the continuous improvement of living standards in modern society, the demand for refined 3D object representation models has grown considerably.Concurrently, the diversification of methods for acquiring 3D point clouds has significantly enhanced our ability to digitally express the surrounding environment.Consequently, the efficient and straightforward generation of high-precision 3D mesh models from point clouds has become particularly vital.As a typical production method for 3D mesh models, point cloud surface reconstruction technology has gained widespread application across various industries owing to its convenience and efficiency.Therefore, in this paper, we propose a distance-constrained scale-space surface reconstruction algorithm, which utilizes the neighborhood distances of point clouds to constrain the anomalous triangular mesh generated by the scale-space reconstruction algorithm, thereby enabling the more precise and efficient surface reconstruction of point clouds.Initially, the raw point set obtained from 3D laser scanning undergoes preprocessing, followed by point cloud reconstruction and mesh construction using the distance-constrained scale-space algorithm.Finally, the constructed mesh is back-projected onto the original point set, resulting in a more refined 3D model.Experiments conducted on the urban underground tunnel dataset successfully reconstructed the surface model of the tunnels, validating the effectiveness of the proposed method and enhancing the visual quality of the model.

Surface reconstruction via cooperative evolutions

Surface reconstruction by parallel and unified particle-based resampling from point clouds

Contact pattern measurement is commonly applied to assess gear transmission and meshing performance in aviation spiral bevel gear (ASBG) manufacturing enterprises. However, these patterns cannot be directly segmented and obtained from point clouds or images because each tooth of the spiral bevel gear has complex 3D spiral surface and texture information. Moreover, research on the high-precision 3D colour reconstruction of spiral gear tooth surfaces is relatively rare. Aimed at these problems, the current work proposed a high-precision 3D colour reconstruction method of spiral bevel gear. This method is based on the fusion of the image and point-cloud information. It includes two core steps to satisfy the practical demands of high precision and high efficiency. First, dense point clouds of the 3D calibration block are converted to a plane using the bird’s eye view method for corner detection. Second, the corners are quickly and accurately detected using a two-stage corner detection algorithm based on the inherent geometric features of the calibration block. The identified corners are then used to solve the transformation matrix from the pixel coordinate system to the line laser coordinate system. Image textures can be precisely projected and fused onto a 3D geometry for the coloured reconstruction of the gear tooth surface on the basis of this transformation matrix. The proposed method was validated by applying it to the 3D colour reconstruction of tooth surfaces of 10 distinct types of ASBGs. The experimental results demonstrated that the re-projection error of the proposed method is less than 0.1 millimetres. The minimal margin of error enables high-precision 3D colour reconstruction of the spiral bevel gear tooth surface. Compared to other classic 3D colour reconstruction methods, the proposed method achieves superior accuracy and efficiency. An experimental device was also developed for the 3D colour reconstruction of the tooth surface of ASBGs, as illustrated in figure 1.

Semantic segmentation of large-scale point clouds by integrating attention mechanisms and transformer models

Structure-From-Motion for Earthwork Planning Khaled Nassar, Ebrahim A.Aly, Younghan Jung Pages 310-316 (2011 Proceedings of the 28th ISARC, Seoul, Korea, ISBN 978-89-954572-4-5, ISSN 2413-5844) Abstract: The reconstruction of surfaces from unorganized point clouds can be of great use in construction applications. Point clouds are mostly created by 3D scanners. Another technique for generating point clouds is the structure-from-motion technique, where the point cloud is generated from a sequence of images. This paper utilizes a surface reconstruction technique to model and quantify earthwork. The described technique can be used in preliminary planning, for project updates, estimating of earthwork quantities, as well as for embedded planning systems in construction equipment. The paper examines the use of structure-from-motion techniques in earthwork application and identifies limitations and, pros, and cons of this approach. Data from 23 earthwork excavation construction sites were collected and analyzed. 3D surface reconstructions were applied on the 23 sites during the construction phases and 3D surfaces were reconstructed and compared to the original land form. In addition, similar experiments were conducted with piles of earth. The results were analyzed to determine appropriate ranges of use for structure-from-motion surface reconstructions in earthwork applications. Results show that the technique is most suited for pit excavations with areas less than 2000 meter squared and with depths up to 5 meters. In addition, piles up to 10 meter highs and with base areas up to 300 meters were successfully reconstructed. The results of this paper should be of great importance to contractors willing to utilize this technology in everyday use. Keywords: Structure-from-motion, Construction, Earthwork, Image Processing, 3D Surface Reconstruction DOI: https://doi.org/10.22260/ISARC2011/0055 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley

It is an important task to reconstruct surfaces from 3D point clouds. Current methods are able to reconstruct surfaces by learning Signed Distance Functions (SDFs) from single point clouds without ground truth signed distances or point normals. However, they require the point clouds to be dense, which dramatically limits their performance in real applications. To resolve this issue, we propose to reconstruct highly accurate surfaces from sparse point clouds with an on-surface prior. We train a neural network to learn SDFs via projecting queries onto the surface represented by the sparse point cloud. Our key idea is to infer signed distances by pushing both the query projections to be on the surface and the projection distance to be the minimum. To achieve this, we train a neural network to capture the on-surface prior to determine whether a point is on a sparse point cloud or not, and then leverage it as a differentiable function to learn SDFs from unseen sparse point cloud. Our method can learn SDFs from a single s parse point cloud without ground truth signed distances or point normals. Our numerical evaluation under widely used benchmarks demonstrates that our method achieves state-of-the-art reconstruction accuracy, especially for sparse point clouds. Code and data are available at https://github.com/mabaorui/OnSurfacePrior.