Search Results (443)

Stereo vision is essential for passive three-dimensional perception in resource-constrained applications that require low power consumption, predictable latency, and explainable geometry. Although deep learning architectures dominate recent benchmarks, the classical block-matching pipeline remains a foundational approach. Optimizing this pipeline involves navigating complex trade-offs among matching robustness, map density, and computational efficiency. This study systematically surveys and physically validates the classical stereo framework. After revisiting geometric first principles, three matching costs (SAD, NCC, ZNCC) are benchmarked alongside Sobel preprocessing and structural refinements, with subsequent validation using a calibrated consumer webcam rig. Middlebury benchmarks (2001–2021) indicate that while SAD fails under complex radiometric distortion, NCC consistently achieves superior quantitative metrics, incurring only a 1.2-fold computational overhead. Extending the disparity search range improves foreground localization, while block size imposes a trade-off between resolving the aperture problem and preserving fine geometric detail. To bridge theoretical analysis and practical deployment, the pipeline is validated using a custom-calibrated consumer stereo rig. The optimized Sobel-NCC architecture is then evaluated for real-time edge deployment on constrained hardware (NVIDIA Jetson Nano) and narrow-baseline sensors (OAK-D SR) in the context of agricultural robotic manipulation. By prioritizing metric precision over dense prediction, the classical pipeline reconstructs target surfaces with approximately 1 cm depth accuracy at 21 frames per second. These results demonstrate that optimized local algorithms offer deterministic and reliable geometric foundations for real-time edge-computed robotics. Although neural networks are essential for dense reconstructions in ill-posed regions, the foundational principles established here remain indispensable for advanced stereo vision system deployment. Full article

Stable three-dimensional hand landmark reconstruction using low-cost RGB-D sensors is important for human–computer interaction, robot teleoperation, and vision-based motion analysis. RGB-based hand landmark detectors provide stable semantic 2D landmarks, but their depth output is not a metric measurement in the physical camera coordinate system. Stereo cameras can provide metric depth, but direct landmark-level back-projection is sensitive to invalid pixels, local depth holes, boundary noise, and partial occlusion. To address these problems, this paper presents a lightweight RGB-D sensing front-end that combines MediaPipe semantic hand landmarks with ZED2 stereo depth. The proposed pipeline detects 21 semantic hand landmarks in the RGB image, obtains landmark-level metric depth from the aligned ZED2 depth map using local median sampling, reconstructs 3D landmarks by camera back-projection, and further applies exponential moving average filtering and a bone-length consistency constraint. Experiments were conducted on a self-collected SVO dataset containing 13 hand actions and 26 recorded sequences, and an additional checkerboard-based reference-distance validation was performed to evaluate the metric depth sampling and 3D back-projection component. Compared with single-pixel sampling, the $5\times 5$ local median strategy slightly increased the valid-depth ratio from 0.9731 to 0.9738 and reduced the temporal smoothness metric from 1.7163 mm to 1.6902 mm. To further justify the temporal filtering choice, an additional comparison with the 1 Euro Filter was conducted using the reconstructed win5 trajectories. The 1 Euro Filter produced stronger smoothing, reducing the temporal smoothness metric to 0.196 mm, but also reduced the path-length ratio to 0.484, indicating substantial motion attenuation. EMA0.7 was therefore retained as a more balanced setting, reducing the temporal smoothness metric to 0.826 mm while maintaining a path-length ratio of 0.803. The BL0.5 bone-length constraint reduced the bone-length standard deviation from 2.0727 mm to 1.1995 mm with limited trajectory modification. The final configuration provides a practical low-cost RGB-D front-end for stable 3D hand landmark reconstruction under controlled indoor conditions. Full article

Background/Objectives: Dynamic navigation and robot-assisted implant workflows depend on robust intraoral perception. Marker-based tracking introduces workflow complexity and is sensitive to occlusions, motivating markerless alternatives. This study evaluates whether a single-stage YOLO instance segmentation model (YOLO-seg) can provide a practical markerless perception layer for dental navigation, combining accurate per-tooth delineation with low, predictable inference latency. Methods: YOLO-seg was trained end to end on an intraoral RGB corpus of 400 training, 20 validation, and 100 testing images, combining a public source and a partner-hospital in-house set. A two-stage YOLO + SAM baseline was implemented for comparison. Segmentation quality was evaluated on a 50-image held-out clinical test set at three complementary levels (per-instance matching, per-class union, and global union), with paired Wilcoxon signed-rank tests, Cliff’s delta effect sizes, and 95% bootstrap confidence intervals. Runtime was assessed under matched inference-only and end-to-end conditions on N = 100 frames at a 640 × 640 resolution on an NVIDIA RTX A2000 GPU. Results: YOLO-seg significantly outperformed YOLO + SAM across all primary metrics, with very large effect sizes (Cliff’s delta: 0.76–0.94; Wilcoxon p < 10⁻⁸ on every metric except precision at IoU ≥ 0.5). YOLO-seg reached AP₅₀ = 0.716 and recall = 0.973 versus 0.383 and 0.398 for YOLO + SAM. Under matched inference-only timing, YOLO-seg ran at 27.08 ms per frame (36.9 FPS) versus 1302.78 ms (0.77 FPS), an approximately 48-fold latency gap intrinsic to the two-stage forward pass. Conclusions: YOLO-seg shows strong potential as a 2D perception module for dental navigation, balancing per-instance segmentation fidelity with real-time feasibility under the tested conditions. These results support its use as a 2D perception front-end for future integration with stereo-based 3D reconstruction and robot-assisted navigation; 3D registration accuracy, implant-placement error, and robotic execution remain outside the scope of the present study. Full article

Dynamic three-dimensional (3-D) reconstruction of small objects moving at high speed is fundamentally limited by the number of viewpoints that a fixed camera array can provide at any single time instant. When the camera count is insufficient, single-frame multi-view stereo produces incomplete or inaccurate geometry. This paper proposes a multi-frame temporal integration approach that overcomes this limitation by exploiting the rigid-body assumption: because a falling object maintains its shape across consecutive frames, images captured at different time instants can be combined into a single, viewpoint-enriched reconstruction. A three-layer circular array of 32 synchronized RGB cameras captures 1440 × 1080 images at 160 fps, and a free-fall-oriented algorithm automatically detects active frames, selects informative temporal windows, and feeds the accumulated multi-frame images into a structure-from-motion and multi-view stereo (SfM-MVS) pipeline, effectively multiplying the number of viewpoints without additional hardware. The algorithm simultaneously recovers the 6-DOF pose trajectory of each object from the SfM-estimated camera parameters. Progressive accumulation experiments on freely falling soybeans (approximately 9–10 mm diameter) show that a single 32-camera frame already achieves an F-score exceeding 0.97 at a 0.5 mm threshold against an industrial structured-light scanner reference, and that accumulating additional temporal frames reaches a stable convergence plateau with both objects reaching a plateau F-score of 0.984. Beyond approximately one to two accumulated frames, additional frames yield diminishing returns, confirming that a small number of temporal frames is sufficient for convergent sub-millimeter accuracy. Across 30 independent free-fall trials with three objects, the system achieves an overall mean error of $0.146\pm 0.033$ mm and an overall F-score of $0.980\pm 0.006$—a mean relative error of approximately 1.6% on 8–10 mm targets—and fine surface features such as structural cracks are resolved at a fidelity sufficient for visual defect identification. These results establish rigid-body multi-frame temporal integration as an effective strategy for high-throughput, non-contact 3-D inspection of small objects in motion. Full article

Pedestrian–vehicle collisions produce a rich kinematic record that is entirely lost by the time a forensic investigation begins. Recovering this record constitutes a state-estimation problem. This paper presents a Phase 1 design for a multimodal sensor fusion and signal-processing framework utilising 128-channel LiDAR, 1080p NIR stereo cameras, and a 2 kHz IMU, all fused via Kalman filtering and Savitzky–Golay polynomial differentiation. The framework is evaluated through Monte Carlo uncertainty propagation and sensitivity analysis applied to a constructed simulation scenario; no real clinical or forensic data are used in this Phase 1 report. Under simulated conditions with throw-distance measurement uncertainty of ±0.5 m, velocity reconstruction shows an estimated propagated uncertainty of ±2.03 km/h under expanded simulation conditions with vehicle-coefficient variance activated. Sensitivity analysis indicates that a 10% noise spike in acceleration would theoretically amplify injury metrics by 26.9%, providing quantitative justification for noise-optimal pre-filtering. The bimodal kinematic–acoustic architecture is proposed as a physically interpretable foundation for collision reconstruction; its experimental performance awaits Phase 2–4 validation. A five-phase validation roadmap is presented, progressing from FEA simulation to independent multi-site replication before any forensic deployment is proposed. Full article

Geomembranes are extensively used for seepage control in the reservoir of pumped-storage power stations due to their superior deformability, ease of construction, and low cost. The deformation behavior of geomembranes under high hydraulic pressure is of great importance for seepage-control design and operational safety evaluation. Nevertheless, existing hydrostatic pressure resistance tests cannot effectively measure the hydraulic bulging deformation of geomembranes subjected to water pressure. This study proposes a non-contact binocular vision method to quantify the hydraulic bulging deformation of geomembranes. The method combines underwater camera calibration, image enhancement, stereo matching, triangulation, and three-dimensional reconstruction to achieve both visualization and accurate measurement of geomembrane deformation. After experimental validation and accuracy calibration, the proposed method was preliminary applied to four geomembrane materials, including HDPE, LLDPE, PVC, and TPO, under hydraulic loading. The results show that the measurement error is less than 5% in the large-deformation range under medium and high water pressures. The method can effectively capture the hydraulic bulging behavior of geomembranes and accurately characterize the deformation features of different materials under high hydraulic pressure. This study provides a practical technical approach for underwater deformation measurement of geomembranes and supports seepage-control design and operational safety monitoring. Full article

Gaussian Splatting is a state-of-the-art technique for 3D reconstruction. In this paper, we investigate how different initialization strategies influence the optimization process within the Gaussian Splatting framework, showing that more accurate initial point clouds can greatly influence the quality of object reconstruction. We introduce the Depth-initialized Gaussian Splatting (DiGS) approach, a pipeline that leverages depth-based initialization. By incorporating depth data from a calibrated stereo camera setup, the proposed method significantly enhances model performance, particularly during the early optimization stages. DiGS is particularly effective for reconstructing isolated single objects and improving the recovery of fine-grained details. Several tests on synthetic and real-world datasets confirm the effectiveness of the proposed pipeline. To evaluate our approach, we employ objective metrics and a user study involving 20 participants to assess with human perception the quality of the proposed approach. Full article

Disparity estimation from stereo-endoscopic images is critical for 3D reconstruction in minimally invasive surgery (MIS). However, surgical environments have inherent interference factors including soft tissue deformation, motion blur, and photometric inconsistency. Currently, self-supervised generative networks such as StyleGAN offer an alternative method, but their reliance on iterative latent optimization leads to high computational latency and limits practical deployment. In this work, we propose a temporal latent prediction method to accelerate this optimization process. Instead of designing a brand new generator, our framework learns to predict an optimized initial latent vector, thereby reducing the number of optimization steps and per-frame inference time. Crucially, this prediction-guided mechanism does not alter the architecture or inference logic of the generator, ensuring the fidelity of reconstruction is comparable to that of the original method. Experiments on Phantom and In vivo datasets demonstrate that our method reduces average optimization steps by 16–59% and cuts per-frame latency by about 2.3×, compared to baseline predictors and initialization strategies. Importantly, the final photometric loss remains nearly identical across all methods, confirming that acceleration does not compromise reconstruction quality. These results position our approach as a practical step toward efficient, self-supervised stereo-endoscopic reconstruction in clinical settings. Full article

Accurate alignment of real-world object poses with their virtual counterparts using sensors, e.g. cameras, is essential for consistent interaction in mixed-reality systems. However, objects can undergo abrupt, untracked movements during periods when a tracking system is inactive, e.g., overnight, causing stored pose records to become inconsistent with the real scene and breaking user interaction in the virtual environment. Off-the-shelf 3D reconstruction networks such as MASt3R (Matching and Stereo 3D Reconstruction) method provide metrically scaled 3D point maps and pixel correspondences, but they are trained on static scenes and therefore fail to produce reliable object correspondences when the object has moved. We propose a robust pipeline that combines MASt3R’s metrically scaled 3D outputs with a background-based alignment strategy to recover and apply the true pose change of moved objects. Our method first segments foreground and background and extracts 3D background point sets for a reference day and a current day. An affine transformation between these background point sets is estimated via a standard registration technique and used to express the current-day object 3D coordinates in the reference coordinate frame. Within that unified frame we compute the object pose change and apply the resulting transform to the virtual object, restoring real–virtual consistency. Experiments on real scenes demonstrate that the proposed approach reliably corrects pose misalignments introduced during inactive periods and substantially improves over applying MASt3R alone, thereby enabling restored and consistent user interaction in the virtual environment. Full article

Surface defects in high-temperature continuous cast billets are critical factors affecting the quality of steel products. Owing to high-temperature radiation, heavy dust contamination, varying billet specifications, and background interference from oxide scales and water stains, existing online surface defect detection technologies for high-temperature continuous cast billets still suffer from limitations including high false-positive rates, inefficient identification of pseudo-defects, and the inability to simultaneously detect three-dimensional (3D) depth information alongside two-dimensional (2D) features. To solve these problems, this paper proposes a multi-dimensional online detection technology for surface defects in high-temperature continuous cast billets based on multi-information fusion. A four-channel multispectral image sensor and a corresponding three-light-source imaging system were developed. Furthermore, a defect sample augmentation method, a deep learning-based 2D recognition method, and a photometric stereo-based 3D reconstruction method were designed to mitigate problems of low detection accuracy and poor robustness caused by sample imbalance among different defect types. Finally, industrial applications were conducted on large-section continuous cast billets, beam blanks, and billets during the grinding process. According to the surface defect detection requirements of different continuous cast billets, multispectral multi-information fusion and traditional 2D defect imaging methods were adopted respectively. The results demonstrate high-precision online detection of surface defects in continuous cast billets, with favorable practical application effects. Full article

Free space detection in assisted driving applications is essential to provide information to vehicles about traversable surfaces and potential obstacles to be avoided. The current trend in free space detection favors the use of deep learning techniques. However, Deep Neural Networks require extensive training that considers as many scenarios as possible, which makes it difficult to create a model that can be generalized to all types of surfaces. Additionally, their lack of explainability contrasts with the growing interest in geometrically grounded and safety-oriented design principles for autonomous vehicle systems. To address these limitations, we propose a geometric approach that incorporates coplanarity conditions and normal vector estimation, removing the dependence on datasets for different types of surfaces. Additionally, the stereoscopic images are clustered in superpixels. The use of images clustered in superpixels allows us to obtain shorter processing times, in addition to taking advantage of the spatial and color information provided by the superpixels to increase the robustness of the three-dimensional reconstruction of the scene. Experimental results show that the proposed superpixel-based approach achieves competitive performance compared to unsegmented dense stereo methods, while significantly reducing algorithmic complexity. These results demonstrate the viability of integrating superpixel clustering into stereo-based free space estimation frameworks. Full article

This manuscript presents a transferable and reproducible methodology for quantitative 3D tree-structure mensuration and transparent, rule-based decision support for tree risk management. The workflow integrates (i) Structure-from-Motion/Multi-View Stereo (SfM–MVS) reconstruction from multi-view imagery, (ii) independent referencing to ensure metric scaling and a consistent local frame, and (iii) point cloud analytics to derive branch-level geometric descriptors (e.g., base diameter, length, inclination, slenderness, and projected reach). A clear rule-based layer operationalizes Tree Risk Assessment Qualification (TRAQ)-style risk components and As Low As Reasonably Practicable (ALARP) principles to map geometry and exposure into auditable management recommendations (e.g., monitoring intervals, pruning/weight reduction, supplemental support, and exclusion-zone planning). To provide a real-data example, the demonstration uses the public Fuji-SfM apple orchard dataset, including three neighboring trees with partially overlapping crowns for tree instance extraction and subsequent TRAQ/ALARP scenarios on an outer tree. The proposed decision layer is intentionally based on external geometry and exposure; internal decay indicators and species-specific mechanical properties (e.g., Modulus of Elasticity (MOE), Modulus of Rupture (MOR)) are outside this demonstration and should be incorporated via complementary diagnostics in operational deployments. Full article

We report the stereoscopic observations of two recurrent streamer waves in a single streamer structure, utilizing coordinated observations from the SOHO, STEREO, and SDO missions. Contrary to the long-held view that fast coronal mass ejections (CMEs) are necessary drivers, we demonstrate that these recurrent waves were excited by two consecutive slow CMEs (<500 km s⁻¹ accompanied by only modest flare activity. Three-dimensional reconstruction reveals that the first and second waves propagated with significant decelerations of −7.93 m s⁻² and −10.26 m s⁻², respectively. Their average amplitudes were $0.41R_{\odot }$ and $0.77R_{\odot }$, wavelengths were $4.02R_{\odot }$ and $6.17R_{\odot }$, and periods were 2.66 and 2.53 h, respectively. While the amplitude of the first wave declined with heliocentric distance (consistent with conventional energy convection), the second wave exhibited an intriguing increasing trend in amplitude. Both waves showed a linear increase in wavelength and period with distance, indicating a non-stationary and dispersive medium. Crucially, despite the disparity in driver energy and wave scales, the periods and their change rates remained nearly identical for both events. This provides compelling case-specific evidence that the streamer wave period is primarily determined by the inherent eigenmodes of the streamer plasma slab rather than the specific characteristics of the trigger. We conclude that the generation of observable streamer waves is a combined consequence of the streamer’s structural stability and the energy transfer efficiency of the triggering disturbance. Full article

Extreme rainfall events can trigger widespread landsliding and fluvial erosion, exerting a disproportionate influence on sediment production and landscape evolution in mountainous watersheds. However, hillslope–channel coupling during individual extreme events remains poorly quantified due to the scarcity of event-scale topographic observations. This study investigates event-scale hillslope–channel coupling by quantifying landslide-driven hillslope erosion and channel incision associated with Typhoon Morakot (2009) in the Sinwulu River watershed, southeastern Taiwan. High-resolution pre- and post-event digital surface models (DSMs) were reconstructed using an aerial structure-from-motion multi-view stereo (SfM–MVS) photogrammetry workflow and corrected for canopy height to derive meter-scale topographic changes. Hillslope and channel domains were delineated, and linked hillslope–channel units were used to examine spatial relationships between erosion processes and topographic and hydraulic factors. Results indicate that landslide erosion dominated sediment production during the event with watershed-average erosion of 544.35 mm, while channel responses exhibited strong spatial contrasts, with pronounced incision in upstream reaches and substantial deposition downstream of major knickpoints. Event-scale analysis provides evidence for a strong correspondence between channel incision and hillslope landslide erosion, whereas correlations with commonly used hydraulic proxies such as unit stream power are comparatively weaker. These findings highlight the value of event-scale topographic measurements for elucidating transient hillslope–channel coupling processes during extreme rainfall events. Full article

Video-based photogrammetric reconstruction is increasingly used when high-resolution still images are unavailable. However, limited spatial resolution, compression artifacts, and motion blur often reduce geometric accuracy. Recent advances in learning-based image super-resolution (SR) offer a promising preprocessing method, but their geometric reliability within photogrammetric workflows remains not well understood. This study provides a controlled quantitative evaluation of learning-based super-resolution for video-based 3D reconstruction. Low-resolution video frames are enhanced using two representative methods: an open-source real-world SR model (Real-ESRGAN ×4) and a commercial solution (Topaz Video AI ×4). All datasets are processed with the same Structure-from-Motion and Multi-View Stereo pipelines and tested against terrestrial laser scanning (TLS) reference data. Results show that super-resolution significantly increases reconstruction density and improves the recovery of fine-scale surface details, while also leading to greater local surface variability compared with reconstructions from the original video; photogrammetric stability remains consistent despite these changes. The findings highlight a fundamental trade-off between reconstruction completeness and local geometric accuracy and clarify when enhanced video imagery via super-resolution can be a reliable source for 3D reconstruction. These results are especially important for spatial data science workflows and AI-powered 3D modeling and digital twin applications. Full article