Search Results (3,794)

Point clouds are widely used to represent real-world objects in the fields of augmented/virtual reality, robotics, etc. However, they are often corrupted with noise, hindering downstream tasks such as mesh surface reconstruction, rendering, etc. In this paper, we revisit sparse point cloud representation for denoising. We observe that existing works generally assume global sparsity of noise across point cloud surfaces, leading to over-smoothing, and typically adopt multiple stages to recover features, inevitably introducing biases and structural inconsistencies. To address these challenges, we propose a structure-aware joint sparse optimization approach for point cloud denoising. Specifically, considering that sparsity primarily lies in feature areas, we propose a novel group-level sparsity-based approximate deviation to characterize surface distortion for different structures. Based on that, we develop an optimization model that can protect structural integrity. We further derive a linear approximate solution and provide a parallel denoising algorithm. Experimental results on different types of datasets demonstrate that the proposed approach outperforms the state-of-the-art works. Full article

Three-dimensional synthetic aperture radar (SAR) point clouds provide valuable geometric observations of urban scenes, but they often suffer from severe noise and layer-like artifacts caused by the low signal-to-noise ratio and tomographic imaging mechanism. These degradations make SAR point cloud denoising significantly more challenging than conventional LiDAR point cloud denoising. In this paper, we propose a Geometry-guided Diffusion SAR Point Cloud Denoising (GDSD) framework to recover geometrically coherent building surfaces from noisy SAR point clouds.The key idea is to exploit relatively clean LiDAR point clouds as geometry priors while avoiding the need for paired SAR–LiDAR supervision or clean SAR ground truth. Specifically, we introduce a Forward Gaussian Noising Process to disrupt the intrinsic layer-like artifacts of SAR point clouds and reduce the input-level discrepancy between SAR and LiDAR domains. We further design a geometry prototype-based alignment module that projects SAR and LiDAR bottleneck features into a shared LiDAR-dominated latent space, enabling geometry-aware conditional reverse diffusion. A DiT-3D-based denoising network is then trained with LiDAR-domain diffusion supervision and applied to SAR point clouds using the aligned SAR geometry condition. To evaluate the proposed method, we construct a SAR point cloud denoising benchmark based on the MV3DSAR dataset with CAD-derived reference surfaces. Experimental results show that GDSD significantly improves the quality of noisy SAR point clouds and clearly outperforms the previous conventional LiDAR point cloud denoising baseline, producing more continuous and geometrically coherent SAR building point clouds. Full article

Robot person following (RPF) in dense crowds requires a mobile robot to maintain an appropriate relative position with respect to a moving target while avoiding surrounding pedestrians and satisfying rear-following and social constraints. This paper proposes PPO-GAT-Follow, an interaction-aware reinforcement learning framework for dense-crowd RPF under geometric visibility loss with available target-relative pose estimates. The follower, target pedestrian, and surrounding pedestrians are represented as graph nodes, and a graph attention encoder models their local interactions. A task-oriented reward mechanism jointly accounts for target maintenance, visibility preservation, collision avoidance, proximity-aware social compliance, rear position maintenance, post-arrival stabilization, and action stability. Experiments are conducted in IR-SIM under fixed-route and random-route settings, with comparisons against MPC, DWA, SFM, and an adapted SARL baseline. In the fixed-route setting with 12 background pedestrians, PPO-GAT-Follow achieves a task success rate of $98.8\%$ and a collision rate of $1.1\%$, improving task success by 10.9 percentage points over MPC. In the random-route setting at the training density, it achieves $83.1\%$ task success and an SPL of $0.815$, outperforming MPC by 18.3 percentage points in task success; at this density, it also surpasses SARL in the main task-level metrics. Zero-shot evaluations across crowd densities, together with structural and reward ablations, reward weight sensitivity analysis, tolerance shift tests, multi-seed training, and stress testing under target pose noise and heterogeneous pedestrian dynamics, further demonstrate the effectiveness and reliability of the proposed framework. Gazebo-based validation also demonstrates system integration feasibility with localization, point cloud-based surrounding pedestrian perception, tracking, and UWB-like target-relative pose input. Nevertheless, visual target identification, re-identification, and perception-level occlusion recovery remain outside the scope of the present validation. Full article

Point cloud perception is important in autonomous driving, robotics, and other security-critical 3D systems, yet learned point cloud classifiers remain vulnerable to transferable adversarial perturbations. A central difficulty in transfer-based black-box attacks is surrogate overfitting: an update that is highly effective on an accessible source model may not generalize to an unknown target architecture. We introduce SpecEOT, a source-agnostic and graph-spectral expectation-over-transformation attack. A fixed graph Fourier transform (GFT) basis is constructed from each clean point cloud. At every optimization iteration, each non-identity view independently samples a frequency band and a perturbation sign from uniform distributions; the resulting view gradients are averaged with equal weights and used to update the adversarial point cloud through projected Adam ascent. We evaluate the stochastic method over repeated seeds, extend the ablation to two source architectures, and analyze the interaction between band count and randomization strength while reporting computational cost and assessing robustness to Gaussian jitter and point dropout. SpecEOT achieves strong transferability on ModelNet40 and ShapeNet. Full article

In recent years, Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) have emerged as promising new approaches for 3D reconstruction. NeRFs rely on neural fields that generate a three-dimensional representation of a scene from photographs, estimating reflectance properties and reconstructing the underlying geometry. Since their introduction in 2020, NeRFs have attracted significant attention due to their wide range of potential applications. Conversely, 3D Gaussian Splatting (3DGS), introduced in 2023, employs Gaussian primitives to efficiently model objects and structures, offering a flexible and adaptive representation of 3D scenes. Starting from an established pipeline for the use of reality-based models for structural analysis, this paper investigates the performance of 3DGS in handling complex geometries and surfaces characterised by challenging acquisition conditions. Full article

Road surface condition monitoring is critical for ensuring safe and efficient transportation networks. This study proposes and evaluates a framework that compares imagery-, Light Detection and Ranging (LiDAR), and accelerometer-based approaches for pavement anomaly detection. The analysis first focused on a 5-mile urban roadway segment, in which all three sensing modalities were evaluated under identical survey conditions using manually interpreted reference anomalies to compare detection accuracy, severity classification, and processing efficiency. The imagery-based Convolutional Transformer-based Crack Segmentation (CT-CrackSeg) model achieved a precision, recall, and F1-score of 88.5%, 88.5%, and 88.5%, respectively, but remained sensitive to environmental factors such as shadows, curbs, roadside features, and pavement texture variations. The LiDAR-based method achieved an F1-score of 93.0%, while the accelerometer-based Isolation Forest and Adaptive Threshold methods achieved F1-scores of 95.2% and 97.2%, respectively. These results indicate strong detection performance under the evaluated validation conditions; however, the reported precision values should be interpreted as dataset-specific rather than universal performance levels. Given the accelerometer-based approach’s strong detection performance, minimal processing time, and low deployment cost, it was further applied across a 36-mile roadway network to evaluate its scalability for network-level monitoring. Across the full route, the spatial agreement among accelerometer systems exceeded 0.91, while the agreement between the two detection methods exceeded 0.96, with 962–996 surface defects detected depending on the sensor and method. Integrating the anomaly detection results into a Potree-based web portal enabled interactive validation with geotagged imagery and point clouds, improving interpretability and diagnostic insight. Overall, the findings highlight that accelerometer-based monitoring, even with consumer-grade sensors, provides a practical, scalable, and low-cost solution for pavement evaluation, while LiDAR and imagery serve as complementary tools for detailed verification and characterization. Full article

Modern autonomous systems rely on heterogeneous sensing modalities—including vision sensors, millimeter-wave radar, and LiDAR—yet each individual sensor exhibits characteristic failure modes in challenging real-world conditions. While LiDAR-vision co-processing has received extensive attention, the synergistic potential of 4D radar paired with monocular optics remains comparatively unexplored. To fill this gap, we develop a graph-enhanced multi-modal architecture that jointly leverages sparse 4D radar returns and high-resolution camera imagery for scene-level 3D perception. The proposed system is organized around four tightly coupled processing stages: (i) an image-guided point densification scheme (SAA) that augments sparse radar clouds with camera-derived pseudo measurements; (ii) a pose-invariant cross-modal fusion layer that harmonizes enriched radar features with image descriptors and object saliency maps; (iii) a dynamic hypergraph assembly stage that captures higher-order inter-object and cross-sensor dependencies; and (iv) a HyperGCN inference module that regresses 3D bounding parameters and class labels on the resulting relational graph. Integrating the temporal velocity cues native to radar with the rich appearance information from cameras, the system delivers reliable perception under diverse environmental conditions. On the View-of-Delft (VOD) evaluation suite, the proposed model records an mAP of 69.3 and mAOS of 59.8. A systematic ablation further quantifies how geometric invariance—across translation, rotation, and scale transformations—individually affects end-to-end detection fidelity. Full article

Background/Objectives: Prostate cancer (PCa) frequently metastasizes to bone, leading to severe clinical complications and reduced quality of life. Accurate and robust imaging-based characterization of bone lesions is therefore critical for diagnosis and treatment planning. Radiomics has emerged as a powerful tool for extracting quantitative information from medical images; however, classical radiomics features are often affected by inter-scanner variability, segmentation dependence, and limited ability to describe lesions with complex biological heterogeneity. This study aims to introduce a translational graph-based radiomics approach designed to extract novel quantitative descriptors with improved robustness and clinical reliability. Methods: A graph representation was derived from segmented Positron Emission Tomography/Computed Tomography (PET/CT) bone lesions by generating a point cloud followed by Delaunay triangulation to preserve geometric information. Graph signal processing techniques were applied to extract three classes of features: orientation, connectivity, and transform-based descriptors. The dataset included PET/CT scans from 50 PCa patients acquired using two different scanners, comprising 92 bone lesions classified as benign or malignant. Correlation analysis with classical radiomics features was performed to assess information redundancy. Robustness against batch effects and segmentation variability was evaluated. Classification performance was tested using Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) models based on proposed features, classical features, and their combination. Results: The proposed features captured non-redundant information compared to classical radiomics and demonstrated superior robustness to scanner-related batch effects and segmentation variability. In classification tasks, models using the proposed features consistently outperformed those based on classical radiomics. Using LDA, the proposed features achieved a mean balanced accuracy of 69.68% and a mean Area Under the Curve (AUC) of 72.14%. With SVM, they achieved a mean balanced accuracy of 65.16% and a mean AUC of 66.49%, exceeding the performance of classical and combined feature sets. Conclusions: This study presents a translational graph-based radiomics framework that extends beyond conventional methodologies, improving robustness and diagnostic performance. The proposed approach shows promise as an integrative tool for more reliable PET/CT-based characterization of bone lesions in prostate cancer. Full article

The flatness of bearing pads directly affects structural load transfer safety. However, conventional total station-based inspection methods suffer from limited spatial sampling, low inspection efficiency, and high safety risks associated with working at height. To address these limitations, this paper proposes UAV-FIBP (Unmanned Aerial Vehicle-based Flatness Inspection for Bridge Pads), an automated and intelligent method for pad flatness assessment utilizing UAV-based 3D reconstruction. By designing a close-range, multi-orbit circumnavigational UAV flight path and acquiring high-overlap imagery (85% forward and 80% side overlap), a millimeter-accurate 3D model is generated via photogrammetry, achieving a high-density point cloud of ≥200 points/cm² on the pad surface. Following point cloud denoising and region-of-interest segmentation using the Random Sample Consensus (RANSAC) algorithm, Principal Component Analysis (PCA) is employed to fit a reference plane. A dual-parameter evaluation framework is subsequently introduced: the root mean square (RMS) deviation quantifies local surface roughness, while the maximum elevation difference is derived from the angle between the normal vectors of the fitted plane and the horizontal plane, thereby enabling a comprehensive assessment of global inclination. Validation experiments conducted on laboratory-scale setups and real construction sites (involving four bridge pads) demonstrate that the proposed method achieves deviations ≤ 2 mm compared to total station measurements, satisfying the requirements stipulated in the Standards for Quality Inspection and Verification of Highways (JTG F80/1-2017). Results indicate that UAV-FIBP enables non-contact, full-coverage, and automated flatness inspection, significantly improving inspection efficiency and construction safety. This work establishes a scalable technical pathway for intelligent bridge construction. Full article

Geospatial Artificial Intelligence (GeoAI) and the rapid advancement of 3D data acquisition technologies (e.g., LiDAR) have enabled scalable semantic interpretation of georeferenced large-scale 3D point clouds across different applications. 3D deep learning-based semantic segmentation on large-scale 3D point clouds typically relies on data partitioning and sampling strategies to address computational constraints, resulting in a substantial proportion of points not being directly predicted by deep neural networks. These points not directly predicted by the model require a processing step for label propagation, which is commonly handled using simple spatial proximity-based rules. While this simplification may be acceptable for some applications, it becomes critical in tasks that require precise spatial measurements and accurate object delineation, where propagation errors can directly affect downstream analyses. To bridge this research gap, this study views this process as a spatial surrogate modeling problem, where predictions from deep learning models are used to infer labels for underrepresented points based on spatial relationships. We adopt inverse distance weighting (IDW) as a transparent, deterministic, and training-free spatial post-processing strategy to examine whether explicitly incorporating spatial relationships improves segmentation outcomes. We evaluate the proposed method within an existing 3D semantic segmentation workflow for bridge inspection, a practical application requiring accurate spatial measurement and reliable object delineation. In the experiments, we use self-collected terrestrial LiDAR point clouds of bridges and associated hydraulic structures and systematically assess how neighborhood size and distance-decay parameters affect model performance on the segmentation task. Results show that this spatial post-processing step provides a modest enhancement over the conventional nearest-neighbor propagation baseline, with notable gains especially on spatially sparse or geometrically complex classes. The results also reveal class-dependent spatial effects, suggesting that different semantic classes exhibit distinct spatial dependencies during label propagation. These findings highlight the practical importance of accounting for spatial context when propagating semantic information to those points not directly predicted by deep neural networks. This is particularly important for downstream applications that require highly accurate spatial measurement and object delineation. The practical value of this post-processing step becomes more apparent in data-scarce application domains such as hydraulic-structure inspection, where large labeled point-cloud datasets and public benchmarks remain limited. In such settings, users may rely on domain-specific pre-trained models, while the original training data may not be publicly available for retraining or extensive model modification. Using a practical case study in the hydraulic domain, this study shows that deterministic post-inference label propagation can improve complete point-wise prediction from an existing model, thereby supporting the reuse of available models for domain-specific applications. The resulting response surfaces support two practical uses: site-specific calibration when limited labeled target data are available, and empirically informed initial settings, a rule of thumb, for comparable bridge-LiDAR applications when target-site labels are unavailable. Full article

Three-dimensional sensing technologies can rapidly acquire 3D point cloud data for object perception and scene understanding. However, point cloud-based object classification is still constrained by limited labeled data and high computational complexity. At present, feature extractors pretrained on large-scale 2D datasets have achieved strong performance in zero-shot 2D classification. Therefore, projecting 3D point clouds into 2D images and leveraging well-established 2D pretrained models for zero-shot point cloud classification has become an effective strategy. In this strategy, generating high-fidelity 2D projections is a critical challenge. To address this challenge, this paper proposes a zero-shot 3D object classification framework based on a multi-scale local geometric feature extraction module and depth-aware multi-view projection. Specifically, point clouds are first modeled as graph structures. Multi-scale radii are used to adjust the receptive field during feature extraction, thereby capturing both fine-grained and large-scale local geometric features. Multi-view 2D images are then generated through depth-wise feature accumulation. These images are fed into a frozen CLIP model for zero-shot classification. The proposed method preserves the structural characteristics of point clouds and reduces the domain gap between point clouds and 2D images. Experiments are conducted on the ModelNet10, ModelNet40, and ScanObjectNN datasets. The results show that the proposed method outperforms current mainstream zero-shot 3D point cloud classification methods. These results suggest that the proposed framework offers an effective solution for point cloud object classification in complex scenarios. Full article

The documentation and management of heritage structures in arid environments present significant challenges due to environmental deterioration, limited historical records, and the complexity of capturing irregular architectural forms. This study presents a structured Scan-to-HBIM workflow integrating terrestrial laser scanning (TLS), photogrammetry, and Heritage Building Information Modelling (HBIM) for the digital documentation of Qasr Harrana, one of the most significant examples of Umayyad desert architecture in Jordan. The proposed workflow combines reality-capture technologies with parametric modelling to generate an information-rich HBIM model that supports the systematic organization of geometric, architectural, and condition-related data. The methodology includes field data acquisition using TLS and digital photography, point cloud processing and registration, photogrammetric image integration, geometric reconstruction, and the development of parametric HBIM components representing key architectural elements of the monument. The resulting model was assessed through geometric verification procedures and was used to document architectural features, spatial organization, and visible deterioration conditions. The study demonstrates how established reality-capture technologies can be integrated within a coherent documentation framework to support the creation of accurate and reusable digital heritage records. Rather than proposing new acquisition algorithms or modelling techniques, the contribution of this research lies in the structured adaptation and implementation of existing Scan-to-HBIM methods for the documentation of desert heritage architecture. The resulting HBIM model provides a comprehensive digital archive that can facilitate future conservation, management, and research activities while improving the accessibility and long-term usability of heritage documentation data. Full article

Traditional manual measurement of garlic bulb phenotypic traits is inefficient, subjective, poorly reproducible, and may cause sample damage. To improve the adaptability of three-dimensional reconstruction to garlic bulb morphology and grading-related parameter extraction, this study developed a non-destructive phenotypic measurement workflow based on multi-view image-based three-dimensional reconstruction. Four garlic materials with distinct bulb morphologies and epidermal characteristics were used to demonstrate the feasibility of the reconstruction workflow, and 40 Lanling white-skinned garlic bulbs were used for quantitative accuracy validation. Multi-view images were acquired using a high-resolution camera, a motorized turntable, and a controlled illumination system. Three-dimensional models were reconstructed using ContextCapture, and the resulting point clouds were processed in CloudCompare through cropping, denoising, downsampling, and pose correction. Maximum longitudinal diameter, maximum transverse diameter, and volume were extracted from the processed point clouds according to GB/T 45244-2025 (Grades and Specifications of Garlic) and validated against manual reference measurements. The coefficients of determination for maximum longitudinal diameter, maximum transverse diameter, and volume were 0.9935, 0.9909, and 0.9924, respectively, with RMSE values of 0.0529 cm, 0.0520 cm, and 0.8874 cm³, and MAPE values of 0.6647%, 0.7765%, and 1.9149%. Additional MAE, bias, confidence interval, and Bland–Altman analyses further supported the agreement between model-derived and manual reference measurements. These results demonstrate the feasibility of multi-view image-based three-dimensional reconstruction for non-destructive garlic bulb phenotypic measurement and provide a methodological basis for future grading-related assessment and three-dimensional phenotyping of bulbous horticultural crops. Full article

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

Rice-milling quality is conventionally inspected post-process, leaving operators unable to correct breakage as it occurs. We present and quantitatively validate a cloud-mediated closed-loop architecture that couples a YOLO26 machine-vision pipeline to Arduino-based actuator control on a laboratory rice mill. The image-acquisition node uploads frames to a cloud repository; an inference and analysis node retrieves them, runs YOLO26 detection with a hybrid post-process classifier to estimate the broken-rice fraction, and issues a command to an Arduino microcontroller that drives PWM-modulated motor and vibrator actuators. The detector achieved a mean Average Precision of 0.951 (peak precision 0.99, peak recall 0.98) on a held-out test set of 100 images. In a matched comparison against an open-loop baseline ($n=196,000$ kernels, broken fraction 21.04%), closed-loop operation ($n=114,000$ kernels) reduced the broken fraction to 6.19%, an absolute improvement of 14.85 percentage points (two-proportion z-test: $z=112.8$, $p<0.001$, 95% CI for the absolute reduction: 14.62–15.08 pp). Dynamic analysis identified a near-linear plant gain of 1.0–1.5% breakage per 1% PWM, providing the empirical basis for future formal PID and Model Predictive Control synthesis. The principal empirical contribution is a quantitative characterisation of the PWM-to-breakage transfer relationship of a rice-milling actuator under deep-learning-derived quality feedback, together with a matched open-loop/closed-loop demonstration that this feedback loop moves the laboratory prototype from non-compliant to Grade A-equivalent quality at constant throughput. The lab-scale prototype is not yet industrial; a roadmap to pilot-scale deployment is outlined. Full article