Search Results (43)

Three-dimensional phenotypic data of pigs provide important information for evaluating growth, nutritional status and health, and they are fundamental to precision feeding, performance assessments, genetic selection and intelligent livestock management. However, point clouds acquired in real pig-house environments are frequently incomplete because of occlusion, limited camera viewpoints, surface reflection, sensor noise, etc. To address local missing structures and geometric discontinuities in pig point clouds, this paper proposes a self-structure-enhanced completion method. The method follows a global-to-local two-stage framework. In the global stage, a self-view fusion network (SVFNet) integrates an incomplete point cloud and its three orthogonal self-projected depth maps to generate a coarse complete shape. In the local stage, a self-structure dual generator (SDG) progressively refines and upsamples the coarse result through a structure analysis and a similarity alignment. To address the physical limitation of distinguishing single-view occlusion from true missingness, this paper proposes a visibility-incompleteness mask (VIM) as the primary contribution, which explicitly models both geometric missingness and multi-view visibility. Furthermore, a prior-adaptive hybrid generation (PAHG) strategy is introduced as a secondary enhancement to combine learnable global shape priors with input-adaptive geometric queries. A dataset containing 1042 complete–incomplete pig point cloud pairs with six typical missing patterns was constructed for model training and evaluation. The proposed method achieved an F-Score@1% of 0.653, a CD-L1 of 9.766, and a CD-L2 of 0.353 on the test set, demonstrating a competitive aggregate performance compared with state-of-the-art completion methods, with trade-offs across different geometric metrics. Full article

Point cloud upsampling is a fundamental task in 3D vision, yet most existing methods adopt a global and uniform strategy, which is computationally inefficient and fails to address the need for region-specific refinement. To address this challenge, we propose PartSPUNet, a novel self-supervised, text-driven point cloud upsampling framework designed to enhance robotic perception through task-oriented local refinement. Inspired by the human cognitive process where high-level language instructions guide visual attention to specific regions of interest, our method allows an operator to use intuitive natural language prompts to direct the upsampling process. Specifically, PartSPUNet leverages a pretrained vision–language model to zero-shot localize the user-specified semantic part within a sparse point cloud. It then performs geometry-aware densification exclusively on this target region, recovering rich geometric details while preserving the global structure. Experimental results demonstrate that our approach significantly outperforms existing methods in reconstructing specified areas, offering a powerful and intuitive tool for enhancing the 3D perception pipeline in intelligent robotic systems. Full article

There are numerous benefits associated with creating digital copies of anatomical structures, which can be used during patient diagnosis. Such models can be used not only for visualization, but also in order to assess the condition of the patient. As advances in both medical imaging and 3D graphics are made, it is necessary to determine areas of application of the known reconstruction algorithms. Specifically, it is crucial to find advantages and disadvantages of known approaches to mesh generation, depending on the properties of the object and compare the quality of their results. In order to provide reliable ground-truth data, three 3D models with features resembling those identified in anatomical structures have been created. Based on these meshes, sets of CT-like DICOM images have been generated. Five different reconstruction approaches were proposed: using 3D occupancy information directly, two ways of obtaining point clouds and two methods that utilize Signed Distance Field. A neural network architecture for the SDF upsampling has also been presented. The obtained results justify the popularity of the Marching Cubes algorithm, as it produced accurate reconstructions most reliably. However, for certain scenarios, promising alternatives have been found. The presented outcomes make it clear that the approach to reconstruction must be tailored to the specific problem. Full article

Collaborative sensing between low-altitude remote sensing and ground-based mobile mapping lays the theoretical foundation for multi-platform 3D data fusion. However, point clouds collected from Airborne Laser Scanners (ALSs) remain scarce due to high acquisition and annotation costs. In contrast, while autonomous driving datasets are more accessible, dense annotation remains a significant bottleneck. To address this, we propose Ground to Altitude (GTA), a weakly supervised domain generalization (DG) framework. GTA leverages sparse autonomous driving data to learn robust representations, enabling reliable segmentation on airborne point clouds under zero-label conditions. Specifically, we tackle cross-platform discrepancies through progressive domain-aware augmentation (PDA) and cross-scale semantic alignment (CSA). For PDA, we design a distance-guided dynamic upsampling strategy to approximate airborne point density and a cross-view augmentation scheme to model viewpoint variations. For CSA, we impose cross-domain feature consistency and contrastive regularization to enhance robustness against perturbations. A progressive training pipeline is further employed to maximize the utility of limited annotations and abundant unlabeled data. Our study reveals the limitations of existing DG methods in cross-platform scenarios. Extensive experiments demonstrate that GTA achieves state-of-the-art (SOTA) performance. Notably, under the challenging 0.1% supervision setting, our method achieves a 6.36% improvement in mIoU over the baseline on the SemanticKITTI → DALES benchmark, demonstrating significant gains across diverse categories beyond just structural objects. Full article

Accurate scene understanding from 3D point cloud data is fundamental to intelligent transportation systems and geospatial digital twins. However, point clouds acquired from lidar sensors in urban environments suffer from incompleteness due to occlusions and limited sensor resolution, presenting significant challenges for precise object localization and geometric reconstruction—critical requirements for traffic safety monitoring and autonomous navigation. To address these point cloud processing challenges, we propose a Center-guided Feature Fusion Completion Network (CFFCNet) that enhances vehicle representation through geometry-aware point cloud completion. The network incorporates a Branch-assisted Center Perception (BCP) module that learns to predict geometric centers while extracting multi-scale spatial features, generating initial coarse completions that account for the misalignment between detection centers and true geometric centers in real-world data. Subsequently, a Multi-scale Feature Blending Upsampling (MFBU) module progressively refines these completions by fusing hierarchical features across multiple stages, producing accurate and complete vehicle point clouds. Comprehensive evaluations on the KITTI dataset demonstrate substantial improvements in geometric accuracy, with localization mean absolute error (MAE) reduced to 0.0928 m and length MAE to 0.085 m. The method’s generalization capability is further validated on a real-world roadside lidar dataset (CUG-Roadside) without fine-tuning, achieving localization MAE of 0.051 m and length MAE of 0.051 m. These results demonstrate the effectiveness of geometry-guided completion for point cloud scene understanding in infrastructure-based traffic monitoring applications, contributing to the development of robust 3D perception systems for urban geospatial environments. Full article

This paper presents a self-tuned two-stage framework for point cloud reconstruction. A parameter-free denoising module (TPDn) automatically selects thresholds through polynomial model fitting to remove noise and outliers without manual tuning. The denoised cloud is then upsampled by PU-Net to recover fine-grained geometry. This synergy enhances structural consistency and demonstrates qualitative robustness under various noise conditions. Experiments on synthetic datasets and real industrial scans show that the proposed method improves geometric accuracy and uniformity while maintaining low computational cost. The framework is simple, efficient, and easily scalable to large-scale point clouds. Full article

Accurate and timely 3D vegetation structure information is essential for ecological modeling and land management. However, these needs often cannot be met with existing airborne LiDAR surveys, whose broad-area coverage comes with trade-offs in point density and update frequency. To address these limitations, this study introduces a deep learning framework built on attention mechanisms, the fundamental building block of modern large language models. The framework upsamples sparse (<22 pt/m²) airborne LiDAR point clouds by fusing them with stacks of multi-temporal optical (NAIP) and L-band quad-polarized Synthetic Aperture Radar (UAVSAR) imagery. Utilizing a novel Local–Global Point Attention Block (LG-PAB), our model directly enhances 3D point-cloud density and accuracy in vegetated landscapes by learning structure directly from the point cloud itself. Results in fire-prone Southern California foothill and montane ecosystems demonstrate that fusing both optical and radar imagery reduces reconstruction error (measured by Chamfer distance) compared to using LiDAR alone or with a single image modality. Notably, the fused model substantially mitigates errors arising from vegetation changes over time, particularly in areas of canopy loss, thereby increasing the utility of historical LiDAR archives. This research presents a novel approach for direct 3D point-cloud enhancement, moving beyond traditional raster-based methods and offering a pathway to more accurate and up-to-date vegetation structure assessments. Full article

Lidar point clouds are crucial for autonomous driving, but their sparsity and scale variations pose challenges for instance segmentation. In this paper, we propose LCPSNet, a Light Detection and Ranging (LiDAR) channel-aware point segmentation network designed to handle distance-dependent sparsity and scale variation in point clouds. A top-down FPN is adopted, where high-level features are progressively upsampled and fused with shallow layers. The fused features at 1/16, 1/8, and 1/4 are further aligned to a common BEV/polar grid and processed by the Local Perception Module (LPM), which applies cross-scale, position-dependent weighting to enhance intra-object coherence and suppress interference. The Inter-Channel Correlation Module (ICCM) employs ball queries to model spatial and channel correlations, computing an inter-channel similarity matrix to reduce redundancy and highlight valid features. Experiments on SemanticKITTI and Waymo show that LPM and ICCM effectively improve local feature refinement and global semantic consistency. LCPSNet achieves 70.9 PQ and 77.1 mIoU on SemanticKITTI, surpassing mainstream methods and reaching state-of-the-art performance. Full article

Point cloud imaging technology usually faces the problem of point cloud sparsity, which leads to a lack of important geometric detail. There are many point cloud upsampling networks that have been designed to solve this problem. However, the existing methods have limitations in local–global relation understanding, leading to contour distortion and many local sparse regions. To this end, PU-DZMS is proposed with two components. (1) the Dense Zoom Encoder (DENZE) is designed to capture local–global features by using ZOOM Blocks with a dense connection. The main module in the ZOOM Block is the Zoom Encoder, which embeds a Transformer mechanism into the down–upsampling process to enhance local–global geometric features. The geometric edge of the point cloud would be clear under the DENZE. (2) The Multi-Scale Complementary Regression (MSCR) module is designed to expand the features and regress a dense point cloud. MSCR obtains the features’ geometric distribution differences across scales to ensure geometric continuity, and it regresses new points by adopting cross-scale residual learning. The local sparse regions of the point cloud would be reduced by the MSCR module. The experimental results on the PU-GAN dataset and the PU-Net dataset show that the proposed method performs well on point cloud upsampling tasks. Full article

The acquisition of high-quality tree point cloud datasets facilitates research in various forestry fields, including tree species classification, diversity monitoring, and biomass estimation. However, due to limitations in sensor performance and occlusion between trees, tree point clouds acquired using LiDAR scanners often exhibit missing data. This not only degrades the quality of the point clouds, but also significantly reduces the number of usable samples. Therefore, this study proposed a tree point cloud enhancement system, which included the completion network and the sample augmentation network. The point cloud completion network utilized a transformer-based improved module to predict missing point clouds and combined up-sampling processing to progressively complete the point clouds from coarse to fine. This could improve the subsequent model decisions and performance through data balancing. On the other hand, the sample augmentation network, based on an adversarial learning strategy, separately constructed the generator and the classifier. By applying shape transformations, point displacements, and point drop to complete point cloud samples, the learnable parameters in the generator and the classifier were alternately optimized. This process enhanced both the quality and the quantity of the tree point cloud dataset. In addition, this study introduced a multi-head attention pooling layer, which further enhanced the joint network’s ability to learn and extract tree structural features. The experimental results showed that the completion network successfully restored missing tree point clouds of various types, achieving an average Chamfer Distance of 4.84 and an average F-score of 0.90. The experiments also demonstrated the effectiveness and robustness of the sample augmentation network, which improved classification accuracy by approximately 2.9% compared to the original dataset. Full article

The increasing reliance on 3D scanning and meshless methods highlights the need for algorithms optimized for point-cloud geometry representations in CAE simulations. While voxel-based binning methods are simple, they often compromise geometry and topology, particularly with coarse voxelizations. We propose an algorithm based on a Series of Local Triangulations (SOLT) as an intermediate representation for point clouds, enabling efficient upsampling and downsampling. This robust and straightforward approach preserves the integrity of point clouds, ensuring resampling without feature loss or topological distortions. The proposed techniques integrate seamlessly into existing engineering workflows, avoiding complex optimization or machine learning methods while delivering reliable, high-quality results for a large number of examples. Resampled point clouds produced by our method can be directly used for solving PDEs or as input for surface reconstruction algorithms. We demonstrate the effectiveness of this approach with examples from mechanically sampled point clouds and real-world 3D scans. Full article

Due to the high-density coverage of vegetation, the complexity of terrain, and occlusion issues, ground point extraction faces significant challenges. Airborne Light Detection and Ranging (LiDAR) technology plays a crucial role in complex mountainous areas. This article proposes a method for constructing fine terrain in high vegetation coverage areas based on implicit neural representation. This method consists of data preprocessing, multi-scale and multi-feature high-difference point cloud initial filtering, and an upsampling module based on implicit neural representation. Firstly, preprocess the regional point cloud data is preprocessed; then, K-dimensional trees (K-d trees) are used to construct spatial indexes, and spherical neighborhood methods are applied to capture the geometric and physical information of point clouds for multi-feature fusion, enhancing the distinction between terrain and non-terrain elements. Subsequently, a differential model is constructed based on DSM (Digital Surface Model) at different scales, and the elevation variation coefficient is calculated to determine the threshold for extracting the initial set of ground points. Finally, the upsampling module using implicit neural representation is used to finely process the initial ground point set, providing a complete and uniformly dense ground point set for the subsequent construction of fine terrain. To validate the performance of the proposed method, three sets of point cloud data from mountainous terrain with different features are selected as the experimental area. The experimental results indicate that, from a qualitative perspective, the proposed method significantly improves the classification of vegetation, buildings, and roads, with clear boundaries between different types of terrain. From a quantitative perspective, the Type I errors of the three selected regions are 4.3445%, 5.0623%, and 5.9436%, respectively. The Type II errors are 5.7827%, 6.8516%, and 7.3478%, respectively. The overall errors are 5.3361%, 6.4882%, and 6.7168%, respectively. The Kappa coefficients of the measurement areas all exceed 80%, indicating that the proposed method performs well in complex mountainous environments. Provide point cloud data support for the construction of wind and photovoltaic bases in China, reduce potential damage to the ecological environment caused by construction activities, and contribute to the sustainable development of ecology and energy. Full article

The upsampling of point clouds is a common task to increase the expressiveness and richness of the details. The quality of upsampled point clouds is crucial for downstream tasks, such as mesh reconstruction. With the rapid development of deep learning technology, many neural network-based methods have been proposed for point cloud upsampling. However, there are common challenges among these methods such as blurring sharper points (e.g., corner or edge points) and producing points gathered together. These problems are caused by similar feature replication or insufficient supervised information. To address these concerns, we present SSPU-FENet, a self-supervised network consisting of two modules specifically designed for geometric detail-preserved point cloud upsampling. The first module, called the feature enhancement module (FEM), aims to prevent feature blurring. This module retains important features such as edges and corners by using non-artificial encoding methods and learning mechanisms to avoid the creation of blurred points. The second module, called the 3D noise perturbation module (NPM), focuses on high-dimensional feature processing and addresses the challenges of feature similarity. This module adjusts the spacing of reconstructed points, ensuring that they are neither too close nor too far apart, thus maintaining point uniformity. In addition, SSPU-FENet proposes self-supervised loss functions that emphasize global shape consistency and local geometric structure consistency. These loss functions enable efficient network training, leading to superior upsampling results. Experimental results on various datasets show that the upsampling results of the SSPU-FENet are comparable to those of supervised learning methods and close to the ground truth (GT) point clouds. Furthermore, our evaluation metrics, such as the chamfer distance (CD, 0.0991), outperform the best methods (CD, 0.0998) in the case of $16\times$ upsampling with 2048-point input. Full article

With the advancement of service robot technology, the demand for higher boundary precision in indoor semantic segmentation has increased. Traditional methods of extracting Euclidean features using point cloud and voxel data often neglect geodesic information, reducing boundary accuracy for adjacent objects and consuming significant computational resources. This study proposes a novel network, the Euclidean–geodesic network (EGNet), which uses point cloud–voxel–mesh data to characterize detail, contour, and geodesic features, respectively. The EGNet performs feature fusion through Euclidean and geodesic branches. In the Euclidean branch, the features extracted from point cloud data compensate for the detail features lost by voxel data. In the geodesic branch, geodesic features from mesh data are extracted using inter-domain fusion and aggregation modules. These geodesic features are then combined with contextual features from the Euclidean branch, and the simplified trajectory map of the grid is used for up-sampling to produce the final semantic segmentation results. The Scannet and Matterport datasets were used to demonstrate the effectiveness of the EGNet through visual comparisons with other models. The results demonstrate the effectiveness of integrating Euclidean and geodesic features for improved semantic segmentation. This approach can inspire further research combining these feature types for enhanced segmentation accuracy. Full article

Although the development of high-resolution remote sensing satellite technology has made it possible to reconstruct the 3D structure of object-level features using satellite imagery, the results from a single reconstruction are often insufficient to comprehensively describe the 3D structure of the target. Therefore, developing an effective 3D point cloud fusion method can fully utilize information from multiple observations to improve the accuracy of 3D reconstruction. To this end, this paper addresses the problems of shape distortion and sparse point cloud density in existing 3D point cloud fusion methods by proposing a 3D point cloud fusion method based on Earth mover’s distance (EMD) auto-evolution and local parameterization network. Our method is divided into two stages. In the first stage, EMD is introduced as a key metric for evaluating the fusion results, and a point cloud fusion method based on EMD auto-evolution is constructed. The method uses an alternating iterative technique to sequentially update the variables and produce an initial fusion result. The second stage focuses on point cloud optimization by constructing a local parameterization network for the point cloud, mapping the upsampled point cloud in the 2D parameter domain back to the 3D space to complete the optimization. Through these two steps, the method achieves the fusion of two sets of non-uniform point cloud data obtained from satellite stereo images into a single, denser 3D point cloud that more closely resembles the true target shape. Experimental results demonstrate that our fusion method outperforms other classical comparison algorithms for targets such as buildings, planes, and ships, and achieves a fused RMSE of approximately 2 m and an EMD accuracy better than 0.5. Full article