Target Point Cloud Research Articles

Lidar Target Point Cloud Alignment Based on Improved Neighborhood Curvature with Iteration Closest Point Algorithm

Lidar Target Point Cloud Alignment Based on Improved Neighborhood Curvature with Iteration Closest Point Algorithm

A Compact In Situ Ranging LiDAR and Performance Evaluation

To capture and process environmental information, a compact lab-built LiDAR is designed and constructed. The lab-built LiDAR utilizes a programmable-scanning-pattern image-space scanner, which is able to satisfy the demands for diverse applications, and a Gauss–Newton (GN) iteration-based in situ ranging system, which is capable of retrieving target distance at high measurement repetition rate. In order to evaluate the angular resolution of the lab-built LiDAR, a trapezoid-fitting method is proposed to quantify the gradually being blurred profile of cubic target, which was validated by the lab-built LiDAR and a commercial LiDAR. Experiments were conducted for the verification of lab-built LiDAR from ranging error, angular error, and angular resolution. The results show that the lab-built LiDAR is able to obtain a target point cloud with a ranging accuracy within [−8, 8] mm, a precision of 4–16 mm, an angular error smaller than 0.26°, and an angular resolution of 6.2 mrad. The proposed LiDAR and evaluation method will benefit hazard detection and avoidance in aerospace applications.

Urban Traffic Imaging Using Millimeter-Wave Radar

Imaging technology enhances radar environment awareness. Imaging radar can provide richer target information for traffic management systems than conventional traffic detection radar. However, there is still a lack of research on millimeter-wave radar imaging technology for urban traffic surveillance. To solve the above problem, we propose an improved three-dimensional FFT imaging algorithm architecture for radar roadside imaging in urban traffic scenarios, enabling the concurrence of dynamic and static targets imaging. Firstly, by analyzing the target characteristics and background noise in urban traffic scenes, the Monte-Carlo-based constant false alarm detection algorithm (MC-CFAR) and the improved MC-CFAR algorithm are proposed, respectively, for moving vehicles and static environmental targets detection. Then, for the velocity ambiguity solution problem with multiple targets and large velocity ambiguity cycles, an improved Hypothetical Phase Compensation algorithm (HPC-SNR) is proposed and complimented. Further, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is used to remove outliers to obtain a clean radar point cloud image. Finally, traffic targets within the 50 m range are presented as two-dimensional (2D) point cloud imaging. In addition, we also try to estimate the vehicle type by target point cloud size, and its accuracy reaches more than 80% in the vehicle sparse condition. The proposed method is verified by actual traffic scenario data collected by a millimeter-wave radar system installed on the roadside. The work can support further intelligent transportation management and extend radar imaging applications.

DOPNet: Achieving Accurate and Efficient Point Cloud Registration Based on Deep Learning and Multi-Level Features.

Point cloud registration aims to find a rigid spatial transformation to align two given point clouds; it is widely deployed in many areas of computer vision, such as target detection, 3D localization, and so on. In order to achieve the desired results, registration error, robustness, and efficiency should be comprehensively considered. We propose a deep learning-based point cloud registration method, called DOPNet. DOPNet extracts global features of point clouds with a dynamic graph convolutional neural network (DGCNN) and cascading offset-attention modules, and the transformation is predicted by a multilayer perceptron (MLP). To enhance the information interaction between the two branches, the feature interaction module is inserted into the feature extraction pipeline to implement early data association. We compared DOPNet with the traditional method of using the iterative closest point (ICP) algorithm along with four learning-based registration methods on the Modelnet40 data set. In the experiments, the source and target point clouds were generated by sampling the original point cloud twice independently; we also conducted additional experiments with asymmetric objects. Further evaluation experiments were conducted with point cloud models from Stanford University. The results demonstrated that our DOPNet method outperforms these comparative methods in general, achieving more accurate and efficient point cloud registration.

HDRNet: High‐Dimensional Regression Network for Point Cloud Registration

AbstractAbstract‐3D point cloud registration is a crucial topic in the reverse engineering, computer vision and robotics fields. The core of this problem is to estimate a transformation matrix for aligning the source point cloud with a target point cloud. Several learning‐based methods have achieved a high performance. However, they are challenged with both partial overlap point clouds and multiscale point clouds, since they use the singular value decomposition (SVD) to find the rotation matrix without fully considering the scale information. Furthermore, previous networks cannot effectively handle the point clouds having large initial rotation angles, which is a common practical case. To address these problems, this paper presents a learning‐based point cloud registration network, namely HDRNet, which consists of four stages: local feature extraction, correspondence matrix estimation, feature embedding and fusion and parametric regression. HDRNet is robust to noise and large rotation angles, and can effectively handle the partial overlap and multi‐scale point clouds registration. The proposed model is trained on the ModelNet40 dataset, and compared with ICP, SICP, FGR and recent learning‐based methods (PCRNet, IDAM, RGMNet and GMCNet) under several settings, including its performance on moving to invisible objects, with higher success rates. To verify the effectiveness and generality of our model, we also further tested our model on the Stanford 3D scanning repository.

A novel point cloud simplification method using local conditional information

In three-dimensional measurement of large-scale parts, such as car bodies and aircraft wings, massive points (up to hundreds of millions) are collected. If all point cloud data is processed, a large amount of computing resources and storage space will be consumed. It is an important task to reduce the burden of processing while maintaining the feature of the point cloud as much as possible. This paper proposes a novel point cloud simplification method using local conditional information that is utilized to evaluate the features of each point in point clouds.The proposed method can reduce the size of point clouds while reserving significant features. Firstly, the original information of each point is evaluated by calculating the curvature of the original point cloud. Secondly, the point with the maximal information is selected and reserved in the target point cloud that will be updated as the alternative point cloud after simplification. The conditional information of each rest point is evaluated by the neighborhood and updated to the information of the current point, which attenuates with the distance from the other reserved points. Finally, the point with the maximal information is reserved iteratively until the number of points reserved satisfies the requirements of simplification according to enterprise needs. Experiments verify the effectiveness and performance of the proposed method. The experimental results indicate that the proposed method performs much better on feature retention than Voxel simplification and Curvature simplification.

A Two-Stage Correspondence-Free Algorithm for Partially Overlapping Point Cloud Registration.

Point cloud registration is a key task in the fields of 3D reconstruction and automatic driving. In recent years, many learning-based registration methods have been proposed and have higher precision and robustness compared to traditional methods. Correspondence-based learning methods often require that the source point cloud and the target point cloud have homogeneous density, the aim of which is to extract reliable key points. However, the sparsity, low overlap rate and random distribution of real data make it more difficult to establish accurate and stable correspondences. Global feature-based methods do not rely on the selection of key points and are highly robust to noise. However, these methods are often easily disturbed by non-overlapping regions. To solve this problem, we propose a two-stage partially overlapping point cloud registration method. Specifically, we first utilize the structural information and feature information interaction of point clouds to predict the overlapping regions, which can weaken the impact of non-overlapping regions in global features. Then, we combine PointNet and the self-attention mechanism and connect features at different levels to efficiently capture global information. The experimental results show that the proposed method has higher accuracy and robustness than similar existing methods.

Attention-Based Transformation from Latent Features to Point Clouds

In point cloud generation and completion, previous methods for transforming latent features to point clouds are generally based on fully connected layers (FC-based) or folding operations (Folding-based). However, point clouds generated by FC-based methods are usually troubled by outliers and rough surfaces. For folding-based methods, their data flow is large, convergence speed is slow, and they are also hard to handle the generation of non-smooth surfaces. In this work, we propose AXform, an attention-based method to transform latent features to point clouds. AXform first generates points in an interim space, using a fully connected layer. These interim points are then aggregated to generate the target point cloud. AXform takes both parameter sharing and data flow into account, which makes it has fewer outliers, fewer network parameters, and a faster convergence speed. The points generated by AXform do not have the strong 2-manifold constraint, which improves the generation of non-smooth surfaces. When AXform is expanded to multiple branches for local generations, the centripetal constraint makes it has properties of self-clustering and space consistency, which further enables unsupervised semantic segmentation. We also adopt this scheme and design AXformNet for point cloud completion. Considerable experiments on different datasets show that our methods achieve state-of-the-art results.

Reliable Inlier Evaluation for Unsupervised Point Cloud Registration

Unsupervised point cloud registration algorithm usually suffers from the unsatisfied registration precision in the partially overlapping problem due to the lack of effective inlier evaluation. In this paper, we propose a neighborhood consensus based reliable inlier evaluation method for robust unsupervised point cloud registration. It is expected to capture the discriminative geometric difference between the source neighborhood and the corresponding pseudo target neighborhood for effective inlier distinction. Specifically, our model consists of a matching map refinement module and an inlier evaluation module. In our matching map refinement module, we improve the point-wise matching map estimation by integrating the matching scores of neighbors into it. The aggregated neighborhood information potentially facilitates the discriminative map construction so that high-quality correspondences can be provided for generating the pseudo target point cloud. Based on the observation that the outlier has the significant structure-wise difference between its source neighborhood and corresponding pseudo target neighborhood while this difference for inlier is small, the inlier evaluation module exploits this difference to score the inlier confidence for each estimated correspondence. In particular, we construct an effective graph representation for capturing this geometric difference between the neighborhoods. Finally, with the learned correspondences and the corresponding inlier confidence, we use the weighted SVD algorithm for transformation estimation.Under the unsupervised setting, we exploit the Huber function based global alignment loss, the local neighborhood consensus loss and spatial consistency loss for model optimization. The experimental results on extensive datasets demonstrate that our unsupervised point cloud registration method can yield comparable performance.

An adaptive grid search algorithm for fitting spherical target of terrestrial LiDAR

The spherical target is critical in many terrestrial laser scanning applications. Accurately and robustly estimating its geometric center based on a point cloud is a problem of widespread concern. To address this problem, we propose a novel adaptive grid search (AGS) algorithm, which makes full use of the point cloud and geometric feature of the spherical target, and obtains the optimal fitting parameters through a finite number of iterative optimizations. Firstly, we utilize the centroid of the point cloud and the estimated radius of the spherical target as constraints to establish an initial constraint space. Secondly, we use grid search to determine the initial optimal solution of the parameters. Finally, we iteratively update the constraint space and the optimal solution utilizing the previous optimal solution and the parameters scales, and take the solution with the least error metric as the final optimal solution. We have validated the feasibility and reliability of the algorithm with actual data and simulated data, respectively, and compared the fitting results with those of the Non-Least Squares algorithm (NLS) and the M−estimator SAmple Consensus (MSAC) algorithm. The experimental results show that the AGS algorithm can achieve high-precision fitting of various point clouds of spherical targets and overcome the defects of the other two algorithms that are easily affected by noise and coverage. Due to iterative optimization, the running time of the AGS algorithm is slightly longer than the other two algorithms, but it can usually complete the fitting within 5 s.

Uniaxial Partitioning Strategy for Efficient Point Cloud Registration.

In 3D reconstruction applications, an important issue is the matching of point clouds corresponding to different perspectives of a particular object or scene, which is addressed by the use of variants of the Iterative Closest Point (ICP) algorithm. In this work, we introduce a cloud-partitioning strategy for improved registration and compare it to other relevant approaches by using both time and quality of pose correction. Quality is assessed from a rotation metric and also by the root mean square error (RMSE) computed over the points of the source cloud and the corresponding closest ones in the corrected target point cloud. A wide and plural set of experimentation scenarios was used to test the algorithm and assess its generalization, revealing that our cloud-partitioning approach can provide a very good match in both indoor and outdoor scenes, even when the data suffer from noisy measurements or when the data size of the source and target models differ significantly. Furthermore, in most of the scenarios analyzed, registration with the proposed technique was achieved in shorter time than those from the literature.

Enhanced Super4PCS Algorithm by Comparing Transformed Normals at Corresponding Points.

In this paper, an enhanced algorithm based on the Super4PCS algorithm was established to address the problem of prolonged congruent set verification of Super4PCS for point clouds with many points or low overlap. By comparing normals of corresponding points in a source point cloud and a tentatively transformed target point cloud, this approach dramatically decreases the time required for candidate transformation verification. This strategy has been shown to improve registration efficiency in experiments.

Estimation of Crop Height Distribution for Mature Rice Based on a Moving Surface and 3D Point Cloud Elevation

Estimation of rice plant height distribution plays a significant role in keeping the feed rate of rice combine harvesters stable. This is an effective way to guarantee the working stability of the whole machine, as a consequence, improving threshing and cleaning efficiency and reducing loss and damage rates. However, dense growth and leafy and bent branches of mature rice make it difficult to detect the lowest point of aggregated growing plants in three dimensional (3D) point cloud data. Therefore, the objective of this study was to put forward a method to estimate plant height distribution on the basis of a moving surface and 3D point cloud elevation. The statistical outlier removal (SOR) algorithm was used to reduce noise points far away from target point cloud body, and then moving surface fitting elevation was applied to achieve accurate classification of ground and crop point cloud data for plant height estimation. Experiments showed that, compared with the actual value, the average square root error (RMSE) of the estimation results was 8.29, the average absolute percentage error (MAPE) was 9.28%, and the average accuracy was 90%. The proposed method could accurately estimate the height of mature rice and is beneficial to calculating the feed rate in advance, which can provide a reference for further investigation in automatic and intelligent harvesting.

Multi‐scale point cloud registration based on topological structure

AbstractThe establishment of point‐to‐point correspondence between point clouds is called three‐dimensional point matching, which is widely used in computer vision, computer graphics, robotics and other fields. When the scale of source point cloud and target point cloud are different, it brings great difficulties to obtain the satisfactory matching result. Therefore, the article proposes a registration algorithm to solve this problem effectively. First, we construct a multi‐scale space related to Gaussian standard deviation. Second, we use a topological similarity of two directed trees constructed by feature points coming from two point clouds for fuzzy matching. Finally, we use local matching and global matching methods to find the corresponding relationship between patches in multi‐scale space. We find the feature correspondence of different scales in multi‐scale space and get the scale factor. Our method improves the accuracy of matching by eliminating mismatch caused by descriptor matching and local geometry matching. We show great performance of our approach by conducting extensive experiments, which outperforms other state‐of‐the‐art methods, in particular, it provides a valuable tool for point cloud registration.

SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration

Point cloud registration is used to find a rigid transformation from the source point cloud to the target point cloud. The main challenge in the point cloud registration is in finding correct correspondences in complex scenes that may contain many noise and repetitive structures. At present, many existing methods use outlier rejections to help the network obtain more accurate correspondences, but they often ignore the spatial consistency between keypoints. Therefore, to address this issue, we propose a spatial consistency guided network using contrastive learning for point cloud registration (SCRnet), in which its overall stage is symmetrical. SCRnet consists of four blocks, namely feature extraction block, confidence estimation block, contrastive learning block and registration block. Firstly, we use mini-PointNet to extract coarse local and global features. Secondly, we propose confidence estimation block, which formulate outlier rejection as confidence estimation problem of keypoint correspondences. In addition, the local spatial features are encoded into the confidence estimation block, which makes the correspondence possess local spatial consistency. Moreover, we propose contrastive learning block by constructing positive point pairs and hard negative point pairs and using Point-Pair-INfoNCE contrastive loss, which can further remove hard outliers through global spatial consistency. Finally, the proposed registration block selects a set of matching points with high spatial consistency and uses these matching sets to calculate multiple transformations, then the best transformation can be identified by initial alignment and Iterative Closest Point (ICP) algorithm. Extensive experiments are conducted on KITTI and nuScenes dataset, which demonstrate the high accuracy and strong robustness of SCRnet on point cloud registration task.

A Representation Separation Perspective to Correspondence-Free Unsupervised 3-D Point Cloud Registration

3D point cloud registration in remote sensing field has been greatly advanced by deep learning based methods, where the rigid transformation is either directly regressed from the two point clouds (correspondences-free approaches) or computed from the learned correspondences (correspondences-based approaches). Existing correspondences-free methods generally learn the holistic representation of the entire point cloud, which is fragile for partial and noisy point clouds. In this paper, we propose a correspondences-free unsupervised point cloud registration (UPCR) method from the representation separation perspective. First, we model the input point cloud as a combination of pose-invariant representation and pose-related representation. Second, the pose-related representation is used to learn the relative pose wrt a "latent canonical shape" for the source and target point clouds respectively. Third, the rigid transformation is obtained from the above two learned relative poses. Our method not only filters out the disturbance in pose-invariant representation but also is robust to partial-to-partial point clouds or noise. Experiments on benchmark datasets demonstrate that our unsupervised method achieves comparable if not better performance than state-of-the-art supervised registration methods.

Using 2-Lines Congruent Sets for Coarse Registration of Terrestrial Point Clouds in Urban Scenes

The registration of terrestrial point clouds is a crucial technique to obtaining the entire scene geometry. However, urban scenes, characterized by repetitive components, symmetrical structures, and object occlusion, pose ambiguity challenge of matches in the accurate registration of terrestrial point clouds. This article proposes a coarse registration approach using 2-lines congruent sets (2LCS) for urban point clouds. 3-D line features are extracted in the point clouds and classified into two categories according to directions. With that, a pair of 3-D line features from different categories is selected in the target point cloud to construct the 2-lines base with significant geometric characteristics. Next, the congruent 2-lines bases are identified in the source point cloud using geometric constraints, leading to the 2LCS. The 2LCS are then used to estimate the rigid motion between pairwise point cloud, leading to a set of candidate transformation parameters. The largest common point (LCP) set and maximum line structure consistency (MLSC) are integrated to evaluate the transformations. Finally, the procedure is repeated based on the RANSAC scheme to achieve optimal transformation values. We applied the proposed method to nine urban scenes with varying conditions and compared its performance with other state-of-the-art methods. The results show that our proposed approach is more robust for registering terrestrial point clouds in urban scenes with repetitive components, symmetrical structures, and limited overlap, leading to optimal or near-optimal results after fine registration optimization.

A Calibration Method for a Self-Rotating, Linear-Structured-Light Scanning, Three-Dimensional Reconstruction System Based on Plane Constraints

This paper proposes a calibration method for a self-rotating, linear-structured-light (LSL) scanning, three-dimensional reconstruction system based on plane constraints. The point cloud of plane target collected by the self-rotating, LSL scanning, 3D reconstruction system should be constrained to the basic principle of the plane equation; it can quickly and accurately calibrate the position parameters between the coordinate system of the LSL module and the coordinate system of the self-rotating, LSL scanning, 3D reconstruction system. Additionally, the transformation equation could be established with the calibrated optimal position parameters. This paper obtains the above-mentioned position parameters through experiments and uses the calibrated self-rotating, LSL scanning, 3D reconstruction system to perform three-dimensional scanning and reconstruction of the test piece. The experimental results show that the calibration method can effectively improve the measurement accuracy of the system.

JoKDNet: A joint keypoint detection and description network for large-scale outdoor TLS point clouds registration

Registration of large-scale outdoor Terrestrial Laser Scanning (TLS) point clouds remains many challenges in the scenes with symmetric and repetitive elements (e.g., park, forest, and tunnel), the weak geometric features (e.g., underground excavation), and dramatically changes in different phases (e.g., mountain). To address these issues, a novel neural network JoKDNet is proposed to jointly learn the keypoint detection and feature description to improve the feasibility and accuracy of point clouds registration. Firstly, a novel keypoint detection module is introduced to automatically learn the score of each sampled point and regard the most significant Top-k sampled points as the detected keypoints. Secondly, an enhanced feature description module is proposed to learn the feature representation of each keypoint by fusing the hierarchical local features and context features. Thirdly, a loss function is designed to make the detected keypoints more distinguishable for matching, which simultaneously maximizes the feature distance between non-corresponding keypoints and minimizes the feature distance of corresponding keypoints. Finally, the distance matrix module and RANdom SAmple Consensus (RANSAC) are utilized to determine the correspondences of source and target point clouds for the transformation calculation. Comprehensive experiments show that the JoKDNet performs effectively on five challenging scenes (e.g., park, forest, tunnel, underground excavation, and mountain) from two datasets (WHU-TLS and ETH-TLS) in terms of registration errors, and robustness to varying scenes, with the maximum rotation error less than 0.06° and maximum translation error less than 0.84 m without ICP.

Virtual Namesake Point Multi-Source Point Cloud Data Fusion Based on FPFH Feature Difference.

There are many sources of point cloud data, such as the point cloud model obtained after a bundle adjustment of aerial images, the point cloud acquired by scanning a vehicle-borne light detection and ranging (LiDAR), the point cloud acquired by terrestrial laser scanning, etc. Different sensors use different processing methods. They have their own advantages and disadvantages in terms of accuracy, range and point cloud magnitude. Point cloud fusion can combine the advantages of each point cloud to generate a point cloud with higher accuracy. Following the classic Iterative Closest Point (ICP), a virtual namesake point multi-source point cloud data fusion based on Fast Point Feature Histograms (FPFH) feature difference is proposed. For the multi-source point cloud with noise, different sampling resolution and local distortion, it can acquire better registration effect and improve the accuracy of low precision point cloud. To find the corresponding point pairs in the ICP algorithm, we use the FPFH feature difference, which can combine surrounding neighborhood information and have strong robustness to noise, to generate virtual points with the same name to obtain corresponding point pairs for registration. Specifically, through the establishment of voxels, according to the F2 distance of the FPFH of the target point cloud and the source point cloud, the convolutional neural network is used to output a virtual and more realistic and theoretical corresponding point to achieve multi-source Point cloud data registration. Compared with the ICP algorithm for finding corresponding points in existing points, this method is more reasonable and more accurate, and can accurately correct low-precision point cloud in detail. The experimental results show that the accuracy of our method and the best algorithm is equivalent under the clean point cloud and point cloud of different resolutions. In the case of noise and distortion in the point cloud, our method is better than other algorithms. For low-precision point cloud, it can better match the target point cloud in detail, with better stability and robustness.