Search Results (487)

The accurate evaluation of gastrointestinal (GI) diseases such as celiac disease (CeD) relies on the assessment of villous architecture, yet progress in imaging-based diagnostics, particularly video capsule endoscopy (VCE), is constrained by the absence of anatomically realistic and reproducible physical models of the intestinal mucosa. Existing benchtop phantoms typically reproduce gross luminal curvature but fail to capture the sub-millimeter villous microstructure, the optical scattering behavior, and the luminal folding of native mucosa that together shape its endoscopic appearance. We developed a modular fabrication framework for an anatomically realistic small intestinal phantom with controlled villous microstructure. High-resolution drop-on-demand photopolymer material jetting was used to print discrete patches of villous-like micropillar arrays with tunable height, diameter, and spacing parameterized from histological data spanning Marsh 0 to 3c classifications. The printed patches were then dyed for mucosal-color realism, bonded onto a polyester–spandex substrate, rolled into a continuous tube, and shaped with adjustable retainer rings to introduce luminal folds. Optical microscopy confirmed dimensional fidelity within ±10% of design values with patch-to-patch variation below 7%, and VCE imaging of healthy and atrophic configurations achieved structural similarity (SSIM) values of 0.625 and 0.761 against clinical mucosal imagery. This reproducible platform supports VCE device validation, imaging dataset generation, and clinician training in gastrointestinal imaging. Full article

Road crack images captured by smartphones suffer from low resolution, uneven illumination, and complex background interference. Mobile devices also have limited resources for real-time high-accuracy segmentation. A two-stage framework combines a high-accuracy dual-path teacher model with a knowledge-distilled lightweight student model. The teacher model integrates a CNN-KAN path for local texture extraction and a VMamba path for global context modeling at linear complexity. A dedicated KAN-based fusion module learns adaptive nonlinear mappings between the two feature streams. On public crack datasets, the teacher model achieves an mIoU of 0.8087 and an mDice of 0.9028. It is then transferred to a self-constructed smartphone dataset built from continuous 30 fps video, where it reaches an mIoU of 0.7084 with strong robustness to illumination and blur. A GAN-based super-resolution strategy further improves the mIoU by 4.01%. A ConvLSTM-based knowledge distillation framework compresses the teacher into a lightweight MobileViT student model. This reduces the parameter count from 57.80 M to 1.57 M and cuts the GPU inference time from 99.56 ms to 1.46 ms, while retaining an mIoU of 0.7078. The deployed student model runs at 15 to 20 frames per second on an Android smartphone. An ablation study confirms that the ConvLSTM-based temporal distillation contributes beyond standard distillation. This framework provides a practical solution for real-time road crack monitoring on smartphones. Full article

This paper presents a hardware–software co-design framework for real-time super-resolution (SR) of low-quality video on resource-constrained edge platforms. At its core is a compact residual network obtained by once-for-all (OFA) neural architecture search over the Residual Channel Attention Network (RCAN) design space, trained conventionally and then optimized with quantization-aware training (QAT) for deployment on an integer-only deep-learning processing unit (DPU). Loop tiling and data-flow scheduling are applied within a custom high-level synthesis (HLS) pre-processing pipeline that feeds the DPU, and a per-directive ablation isolates the contribution of each optimization to post-route resource usage and timing. Deployed on a Kria KV260 board with a $128\times 128$ network input, the INT8 network sustains $96.37$ FPS at the ×2 scale at a measured board power of $5.38$ W, corresponding to $6.32$ Mpixel/s of reconstructed output at $1.17$ Mpixel/J, within 63.2% of the device LUT budget and with timing closed at 275 MHz. Relative to the FP32 model, INT8 quantization costs $0.274$ dB of peak signal-to-noise ratio (PSNR) on Set5, $0.172$ dB on Set14, $0.116$ dB on B100, and $0.146$ dB on Urban100, a loss dominated (81–90%) by activation rather than weight quantization. On a held-out UAV subset drawn from VisDrone2019, which is the operating domain the system targets, the network reconstructs at $25.94$ dB and $0.748$ SSIM. These results show that a twenty-three-layer residual SR network can be deployed within a 5.38 W envelope on a low-cost integer-only edge FPGA, making the approach suitable for autonomous systems, robotics, and airborne surveillance. Full article

UAV-based ship segmentation is important for berthing and unberthing monitoring, ship–berth distance estimation, and situational awareness in port waters. However, direct video mask propagation with Segment Anything Model 2 (SAM 2) remains susceptible to local contour degradation in high-resolution UAV videos containing weak berth-side boundaries, adjacent tugboats, quay-side structures, water-surface reflections, and target-scale variations. To address this problem, a constrained local boundary refinement method is proposed for target-ship segmentation. The method follows a training-free, first-frame-mask-initialized semi-supervised video object segmentation setting, with all SAM 2 parameters remaining frozen. ROI Boundary Re-Inference first enhances weak contours within local target neighborhoods. Prompt-Consensus Refinement then retains boundary candidates consistently supported by multiple structured prompt variants. Finally, Boundary-Constrained Non-Erosive Fusion restricts supplementation to a narrow neighborhood of the propagated boundary and incorporates reliable candidates without deleting the original foreground. Experiments on a self-built UAV berthing and unberthing video dataset show that the proposed method improves Boundary F@5 px from 89.68% to 94.03% and J&F from 94.08% to 96.39%. These results demonstrate that the proposed method improves target-ship boundary delineation without model training or fine-tuning. Full article

Rapid global urbanization is increasing the need for accurate, large-scale traffic monitoring to support sustainable transportation and city governance. Satellite video remote sensing offers a unique way to continuously observe urban road networks over large areas. It provides high-resolution spatio-temporal data that is essential for traffic flow analysis, infrastructure assessment, and dynamic urban planning. Moving vehicle detection in satellite video sequences is a basic task that turns raw imagery into useful traffic-state information, supporting these applications. Despite the advantages of satellite video data, detecting moving vehicles in practice remains a tough problem. Objects are extremely small and lack clear appearance details, while low local contrast makes them hard to separate from complex backgrounds. Satellite platform motion also introduces background misalignment and intensity fluctuations, resulting in missed detections and false alarms that hurt monitoring reliability. Furthermore, current methods do not fully exploit temporal motion cues or transform-domain priors, creating a performance bottleneck that restricts their practical use. To solve these problems, this paper proposes a Channel-Attentive Spatio-Temporal-Frequency Alignment (CASTFA) framework to effectively use and combine multi-dimensional features for moving vehicle detection in satellite videos, with the goal of providing high-quality traffic monitoring data to help smart city planning. Specifically, a State Space-Guided Temporal Compression (SSGTC) module first collects information along the time dimension with linear computational complexity, greatly reducing overhead while keeping motion cues that are critical for traffic-state estimation. The compressed temporal features are then processed with a multi-scale Haar wavelet transform to get hierarchical time-frequency representations that capture subtle motion dynamics across different frequency bands. At the same time, a pre-trained backbone network extracts multi-scale spatial features. To allow these different domains to work together, a Cross-Domain Feature Alignment (CDFA) mechanism aligns and combines spatial and time-frequency features through channel-attentive operations. Experimental results on the publicly available satellite video moving vehicle detection dataset show that the proposed CASTFA method consistently outperforms existing approaches, with better precision, recall, and F1-scores across diverse urban scenarios. These results show that CASTFA can provide reliable moving vehicle detection performance under difficult real-world conditions, supporting accurate traffic-flow monitoring and providing valuable geospatial intelligence for smart urban planning, transportation management, and sustainable city development. Full article

Multimodal large language models (MLLMs) face substantial memory bottlenecks when processing long visual contexts, such as videos and high-resolution images. Existing methods that allocate visual KV cache budgets using cross-modal attention entropy mainly estimate the distributional breadth of text–vision interaction and may overlook how visual relevance varies across query positions in the encoded multimodal context. We propose a cross-modal query-position variance-aware KV cache optimization method for efficient multimodal long-context inference. The proposed method combines cross-modal attention entropy with prefill-stage query-position variance computed from cross-modal attention to estimate layer-wise visual KV cache preferences. Based on this preference score, visual KV cache budgets are allocated across layers, and variance-aware token pruning is applied to retain high-importance KV states while directly evicting redundant visual tokens without feature merging. Experiments on the MileBench benchmark using LLaVA-v1.5-7B show that, under deterministic single-run evaluation and while retaining only 20% of the visual KV cache, the proposed method produces point-estimate performance close to the full-cache reference and higher point estimates on several fine-grained reasoning and retrieval subtasks. Additional representative-subtask evaluations under different visual cache budgets and on InternVL2.5-8B further provide preliminary point-estimate evidence that the proposed allocation signal is not restricted to a single cache ratio or backbone. System profiling further shows that the 20% cache setting reduces measured KV cache GPU memory from 1.28 GiB to 0.26 GiB and decoding latency from 100.28 ms/token to 92.85 ms/token. These results suggest that cross-modal query-position variance may help preserve sparse, query-dependent visual cues under low-cache-budget multimodal inference. Full article

This paper presents a lightweight MCU–EdgeTPU platform—a microcontroller unit (MCU) paired with an Edge Tensor Processing Unit (EdgeTPU) accelerator—for onboard drone-perception experiments, extended from an open-source baseline originally limited to Quarter Video Graphics Array (QVGA) single-camera operation. Rather than treating hardware, runtime, model, and data as separate problems, they are developed as parts of the same continuous perception pipeline. The platform extends the hardware baseline toward dual 5 Mpx sensing, onboard inertial measurement unit (IMU) support, real-time embedded inference, and a high-level MicroPython control layer. In parallel, lightweight You Only Look Once (YOLO) detectors are trained and selected on a synthetic aerial-person dataset generated under the visual conditions expected by the drone camera, including target resolution, viewpoint, object scale, weather, lighting, and time-of-day variation. The resulting workflow starts from both ends: the detector must be small and quantization-stable enough for the EdgeTPU path, while the dataset must match the images that the onboard sensor is expected to observe. To evaluate the system, the full path from camera capture and image conversion to TPU transfer, model execution, and post-inference processing is analyzed. In the tested setup, the optimized single-camera pipeline runs stably with no timeouts or inference failures at about 26 detections per second with standard RGB input; because each EdgeTPU invocation is bounded by the USB transfer of the input image, feeding the camera’s native YUV420 format instead halves that transfer and raises throughput to about 40 detections per second at the same accuracy, while the selected 8-bit-integer (INT8) person detector preserves most of its 32-bit floating-point (FP32) accuracy. Detections are exposed to drone-control workflows (MAVLink/PX4 and Crazyflie) through the scriptable layer as an integration interface rather than a validated autonomy stack. The central contribution is therefore a co-designed embedded perception pipeline in which the board, runtime, detector, dataset, and even the camera pixel format are aligned around the same operating conditions. Full article

To address the challenges of the large-scale georeferencing of tower-based cameras and the limited capability of video-based spatial analysis, we proposed a geospatial mapping method integrating 3D GIS and gradient descent optimization. Using a Digital Elevation Model (DEM), high-resolution remote sensing imagery, and tower-based video data as the primary data sources, the proposed method first estimates the intrinsic parameters of the tower-based camera by aligning a 3D GIS virtual camera with the video imagery. Subsequently, the initial camera extrinsic parameters are estimated using the PnP algorithm based on the previously estimated intrinsic matrix $K$ and the corresponding control point pairs. Building upon these initial estimates, the camera intrinsic and extrinsic parameters are jointly optimized using a constrained L-BFGS-B framework that incorporates prior knowledge of the tower planar location, explicit box constraints, and a semi-constrained parameterization scheme with bounded parameter ranges. Furthermore, an outlier-removal and re-optimization strategy is employed to further improve the accuracy of parameter estimation. Finally, the optimized parameters are employed to transform image coordinates into three-dimensional world coordinates, and video geospatial mapping is achieved through the integration of colored point clouds with the 3D GIS scene. The results showed the following: (1) The 3D GIS scene constructed from publicly available DEM and high-resolution remote sensing imagery met the requirements for the initial estimation of intrinsic and extrinsic camera parameters. (2) Compared with PnP, RANSAC-PnP, SQPnP, and DLT, the proposed method achieves lower reprojection and 3D spatial errors. For the independent check points, the RMSE of the reprojection error is reduced by 66.4%, 73.6%, 68.0%, and 48.3%, respectively, while the RMSE of the 3D spatial error is reduced by 84.6%, 86.2%, 83.1%, and 69.4%, respectively. These results demonstrate that the proposed method provides reliable camera parameter estimates for video geospatial mapping. (3) Using the estimated camera parameters, image coordinates are transformed into 3D world coordinates to generate a georeferenced colored point cloud, which facilitates integrated analysis with existing geospatial datasets. The proposed method provides a feasible solution for tower-based camera georeferencing and three-dimensional visualization under conditions without field calibration. It offers a theoretical and technical basis for geospatial monitoring and related applications. Full article

This work offers an extensive performance evaluation of video-based pothole detection algorithms utilizing a unique dataset of 619 high-resolution movies recorded in South Kalimantan, Indonesia. Seven distinct models were assessed: three multi-frame-based methodologies (Best Frame Selection, Temporal Consistency Loss, and Multi-Frame Ensemble) employing U-Net architectures with temporal modeling, three per-frame models (OneFormer, YOLOv8-seg, and YOLACT), and one fusion ensemble integrating the per-frame models via weighted boxes fusion. The video collection consists of 2 s segments containing 48 frames each, accompanied by ground truth segmentation masks for pothole identification. Results indicate that per-frame models substantially surpass video-based methods, with the fusion ensemble attaining 81% IoU, followed by YOLOv8-seg and OneFormer, each getting 80% IoU. Parameter efficiency investigation indicates that YOLOv8-seg is the most efficient, achieving IoU per million parameters. Full article

Passenger car unit (PCU) values are an essential input for traffic capacity assessment (TCA) of intersections, as they allow different vehicle categories to be converted into a common unit. In the Slovak Republic, current technical guidelines use fixed equivalency factors for specific intersection types. However, international research shows that PCU values depend on local traffic conditions, vehicle composition, road geometry, and vehicle interactions. Incorrectly selected factors may therefore lead to inaccurate capacity calculations and misleading conclusions regarding intersection performance. This study analyses PCU values for different vehicle categories, with a focus on heavy vehicles (HV) at roundabouts and turbo roundabouts (TR). Traffic surveys were conducted at selected intersections near industrial areas, where a higher proportion of freight traffic was expected. Manual and semi-automatic turning-movement counts were combined with high-resolution video recordings and automatic traffic counters (ATC) to obtain data on traffic volumes, vehicle composition, travel times, speeds, vehicle lengths, and time headways. The results indicate that the behavior of trucks and HV combinations may differ from the assumptions reflected in static equivalency factors. In several cases, the measured travel times and time headways did not reach the values implied by the prescribed PCU coefficients. Based on these findings, a methodology for dynamically determining PCU values was proposed. The methodology is based on the time headway principle and uses commonly available measurement devices. The proposed approach enables PCU values to be determined for either a simplified two-category vehicle classification or a more detailed classification. It may serve as an alternative to static tabulated values, particularly under non-standard traffic composition, a high proportion of HV, or specific geometric conditions of intersections. Full article

Deep multiscale learning has emerged as a promising venue for robust image detection and multi-object tracking in adverse conditions, but the current solutions tend to be impacted by the issues of occlusion, scale variation, and background clutter, focusing on each of them separately and restricting the generalization. In a direction to address these gaps, this piece of writing proposes a unified model that incorporates HRNet to extract high-resolution features, DETR to make use of transformers for detection, and TrackFormer to identify in an identity-preserving manner. Data was based on the MOT17 benchmark dataset, which provides various urban video sequences, including annotated bounding boxes and identities, to guarantee a test that is rigorous. The approaches were selected due to their complementary advantages: HRNet keeps fine-grained spatial information, DETR allows us to locate the objects in an accurate way, and TrackFormer tracks the trajectories across fragments. Experiments show good performance, with a mean detection AP of 70.9, precision of 76.5, recall of 72.8, MOTA of 74.8, IDF1 of 70.2, and HOTA of 63.6, maintaining real-time performance of 26 FPS with a latency of 38.5 ms per frame. In general, this work offers a globally scalable, end-to-end system for problems like surveillance and self-driving, and future work aims to address outrageously dense scenes, enhance cross-dataset generalization, and come up with lightweight systems to deploy these edges. Full article

Crime surveillance in urban environments faces increasing challenges due to dynamic conditions and the demand for real-time monitoring. This paper investigates the use of video data from autonomous vehicles to enhance situational awareness in public spaces through deep learning models optimized for edge processing. High-resolution vehicle-mounted cameras serve as mobile surveillance units capable of real-time object detection, human action recognition, and anomaly detection, bridging the gap between autonomous mobility and urban monitoring. Building on this vision, we introduce ITS-Vision, a generic framework that operationalizes these use cases, enabling autonomous vehicles to function as mobile, context-aware sensing platforms. To validate this approach, we develop prototypes for key ITS-Vision components: a fight detection module using a fine-tuned X3D model, suspect identification via MediaPipe for detection combined with FaceNet for embedding extraction, and a dangerous items detection module using a fine-tuned YOLOv11n model. Due to the limited availability of real-world autonomous vehicle video datasets, experiments were conducted in controlled laboratory environments, demonstrating the feasibility of the proposed architecture and algorithms under simulated conditions. Future work will focus on collecting dedicated datasets and advancing the models toward deployment in real urban scenarios. Full article

In recent years, unmanned aerial vehicles (UAVs) and Big Data analytics have both emerged as increasingly important approaches in advanced traffic monitoring. UAVs provide high-resolution spatial data and operational flexibility, supporting automated vehicle detection and the construction of origin–destination (O/D) matrices through video processing. Conversely, Big Data offers a passive and non-invasive approach based on heterogeneous sources such as mobile devices, satellite navigation systems, and digital applications, ensuring continuous temporal coverage for mobility pattern analysis. This study evaluates the combined use of UAVs and Big Data for traffic flow monitoring as an alternative to traditional manual methods. Focusing on two case studies in Trapani (Italy), the research assesses the advantages and limitations of each technology and their complementary use. Results show that Big Data effectively captures large-scale temporal dynamics but lacks accuracy for detailed O/D estimation, while UAVs provide precise spatial and behavioural information despite operational constraints. A key objective of this study is to investigate the potential complementarity between UAV observations and Big Data traffic monitoring technologies, highlighting the main strengths and limitations of each method under complex study sites and challenging operational conditions for traffic data acquisition using UAVs. Full article

This study presents the diagnostic and conservation work carried out on the Roman mosaic of the South cubiculum in the Latomia dei Niccolini (Marsala, western Sicily). The mosaic, decorated with polychrome tesserae featuring a kantharos motif, presented severe structural damage, including fractures, subsurface voids, and progressive material loss. To assess the causes of deterioration and design an effective conservation strategy, an integrated approach combining non-invasive geophysical and 3D survey methods was applied. Ground-penetrating radar (GPR) was selected as the main diagnostic tool because it allows high-resolution subsurface imaging while preserving the integrity of the fragile mosaic surface. By utilizing high-frequency 2 GHz antennas and complementary video inspection, a significant subsurface cavity beneath the mosaic preparation layer was successfully mapped, determining its critical relationship with the main diagonal surface fracture. Simultaneously, laser scanning and close-range photogrammetry enabled the creation of accurate 3D models supporting both documentation and restoration planning. The conservation concluded with surface cleaning, mortar consolidation, and the successful structural detachment and relocation of the compromised section onto a lightweight support for future museum display. The findings demonstrate that integrating 3D digital and geophysical data provides a quantitative, low-risk roadmap for preserving highly vulnerable archaeological floorings, moving beyond qualitative technical documentation to establish a replicable preservation framework. Full article

Background/Objectives: Single-case studies represent a sophisticated and rigorous methodological framework, widely established in clinical research for providing high-resolution data on individual functional responses. This study evaluates the clinical utility of integrating immersive Virtual Reality (VR) gaming as a novel “functional ingredient” within the Video Game Therapy (VGT) protocol. Given the exploratory single-case nature of this intervention, clinical state-modulations cannot be rigorously validated using standard aggregated group statistics. Therefore, the core objective of this paper is to investigate the therapeutic potential of the VR session on psychological state-modulation, introducing the Single-Case Normative Analytics (SCNA) framework as the mandatory statistical vehicle required to validate individual longitudinal shifts against normative data. Methods: The study treats individual VR exposures as independent, short-term clinical probes embedded within a real-world clinical journey. The SCNA framework was deployed by integrating Crawford’s modified t-tests with longitudinal percentile tracking against an empirical normative reference group (n = 20). Acute state-anxiety variations (STAI-Y1), psychological well-being (PGWBI), and flow dynamics were tracked across three distinct sessions to monitor the patient’s relative repositioning within the normative distribution. Results: The inferential analysis indicates that the immersive 20-min environment facilitated reliable, statistically significant changes in acute state anxiety and flow dimensions, systematically exceeding standard measurement error boundaries and successfully moving the patient’s psychometric profile toward healthy normative ranges. Conclusions: While these findings focus on individual, idiographic reactivity, they demonstrate the utility of the SCNA framework in providing clinicians with objective, evidence-based feedback on the clinical viability of specific VR-based functional units. This approach allows for a rigorous evaluation of standalone digital tools independently of a full, holistic VGT protocol, offering a structured alternative to traditional designs focused on identifying general patterns across groups. Full article