Search Results (750)

Cracks in civil infrastructure exhibit large variations in appearance due to differences in surface texture, illumination, and background clutter, making reliable segmentation a challenging task. To address this issue, this paper proposes an adaptive Mixture-of-Experts (MoE) framework that combines multiple crack segmentation models based on their estimated reliability for each input image. A lightweight gating network is trained using out-of-fold soft supervision to learn how to rank and select the most suitable experts under varying conditions. During inference, only the top two experts are combined to produce the final segmentation result. The proposed framework is evaluated on two public datasets—Crack500 and the CrackForest Dataset (CFD)—and one in-house dataset (RCFD). Experimental results demonstrate consistent improvements over individual models and recent state-of-the-art methods, achieving up to 2.4% higher IoU and 2.1% higher F1-score compared to the strongest single expert. These results show that adaptive expert selection provides an effective and practical solution for robust crack segmentation across diverse real-world scenarios. Full article

Daily nighttime light (NTL) observations made by remote sensing satellites can monitor human activity at high temporal resolution, but are often constrained by residual physical disturbances. Even in standard products, such as NASA’s Black Marble VNP46A2, factors related to sensor viewing geometry, lunar illumination, atmospheric conditions, and seasonality can introduce noise into daily radiance retrievals. This study develops a locally adaptive framework to diagnose and correct residual disturbances in daily NTL data. By estimating location-specific regression models, we quantify the residual sensitivity of VNP46A2 radiance to multiple disturbance factors and selectively remove statistically significant components. The results show that the proposed approach effectively removes statistically significant residual disturbances from daily NTL data in the VNP46A2 product. An application for COVID-19 containment periods in China demonstrates the effectiveness of the proposed approach, where corrected daily NTL data exhibit enhanced temporal stability and improved interpretability. Further analysis based on event study approaches demonstrates that corrected daily NTL data enable the identification of short-run policy effects that are difficult to detect with lower-frequency indicators. Overall, this study enhances the suitability of daily NTL data for high-frequency socioeconomic applications and extends existing preprocessing approaches for daily NTL observations. Full article

Computer vision and deep learning are essential in diverse fields such as autonomous driving, medical imaging, face recognition, and object detection. However, enhancing low-light remote sensing images remains challenging for both research and real-world applications. Low illumination degrades image quality due to sensor limitations and environmental factors, weakening visual fidelity and reducing performance in vision tasks. Common issues such as insufficient lighting, backlighting, and limited exposure create low contrast, heavy shadows, and poor visibility, particularly at night. We propose QWR-Dec-Net, a quaternion-based Retinex decomposition network tailored for low-light image enhancement. QWR-Dec-Net consists of two key modules: a decomposition module that separates illumination and reflectance, and a denoising module that fuses a quaternion holistic color representation with wavelet multi-frequency information. This structure jointly improves color constancy and noise suppression. Experiments on low-light remote sensing datasets (LSCIDMR and UCMerced) show that QWR-Dec-Net outperforms current methods in PSNR, SSIM, LPIPS, and classification accuracy. The model’s accurate illumination estimation and stable reflectance make it well-suited for remote sensing tasks such as object detection, video surveillance, precision agriculture, and autonomous navigation. Full article

Ellipsometric mapping measurements and Bayesian evaluation were performed with a non-collimated, imaging ellipsometer using an LCD monitor as a light source. In such a configuration, the polarization state of the illumination and the local angle of incidence vary spatially and spectrally, rendering conventional spectroscopic ellipsometry inversion methods hardly applicable. To address these limitations, a multilayer optical forward model is augmented with instrument-specific correction parameters describing the polarization state of the monitor and the angle-of-incidence map. These parameters are determined through a Bayesian calibration procedure using well-characterized Si-SiO₂ reference wafers. The resulting posterior distribution is explored by global optimization based on simulated annealing, yielding a maximum a posteriori estimate, followed by marginalization to quantify uncertainties and parameter correlations. The calibrated correction parameters are subsequently incorporated as informative priors in the Bayesian analysis of unknown samples, including polycrystalline–silicon layers deposited on Si-SiO₂ substrates and additional Si-SiO₂ wafers outside the calibration set. The approach allows consistent propagation of calibration uncertainties into the inferred layer parameters and provides credible intervals and correlation information that cannot be obtained from conventional least-squares methods. The results demonstrate that, despite the broadband nature of the RGB measurement and the limited number of analyzer orientations, reliable layer thicknesses can be obtained with quantified uncertainties for a wide range of technologically relevant samples. The proposed Bayesian framework enables a transparent interpretation of the measurement accuracy and limitations, providing a robust basis for large-area ellipsometric mapping of multilayer structures. Full article

The widespread application of robots in complex and dynamic environments demands that Visual SLAM is both robust and accurate. However, dynamic objects, varying illumination, and environmental complexity fundamentally challenge the static world assumptions underlying traditional SLAM methods. This review provides a comprehensive investigation into the mathematical foundations of V-SLAM and systematically analyzes the key optimization techniques developed for dynamic environments, with particular emphasis on advances since 2020. We begin by rigorously deriving the probabilistic formulation of V-SLAM and its basis in nonlinear optimization, unifying it under a Maximum a Posteriori (MAP) estimation framework. We then propose a taxonomy based on how dynamic elements are handled mathematically, which reflects the historical evolution from robust estimation to semantic modeling and then to deep learning. This framework provides detailed analysis of three main categories: (1) robust estimation theory-based methods for outlier rejection, elaborating on the mathematical models of M-estimators and switch variables; (2) semantic information and factor graph-based methods for explicit dynamic object modeling, deriving the joint optimization formulation for multi-object tracking and SLAM; and (3) deep learning-based end-to-end optimization methods, discussing their mathematical foundations and interpretability challenges. This paper delves into the mathematical principles, performance boundaries, and theoretical controversies underlying these approaches, concluding with a summary of future research directions informed by the latest developments in the field. The review aims to provide both a solid mathematical foundation for understanding current dynamic V-SLAM techniques and inspiration for future algorithmic innovations. By adopting a math-first perspective and organizing the field through its core optimization paradigms, this work offers a clarifying framework for both understanding and advancing dynamic V-SLAM. Full article

Self-supervised monocular depth estimation has demonstrated high practical utility, as it can be trained using a photometric image reconstruction loss between the original image and a reprojected image generated from the estimated depth and relative pose, thereby alleviating the burden of large-scale label creation. However, this photometric image reconstruction loss relies on the Lambertian reflectance assumption. Under non-Lambertian conditions such as specular reflections or strong illumination gradients, pixel values fluctuate depending on the lighting and viewpoint, which often misguides training and leads to large depth errors. To address this issue, we propose a multitask learning framework that integrates albedo estimation as a supervised auxiliary task. The proposed framework is implemented on top of representative self-supervised monocular depth estimation backbones, including Monodepth2 and Lite-Mono, by adopting a multi-head architecture in which the shared encoder–decoder branches at each upsampling block into a Depth Head and an Albedo Head. Furthermore, we apply Intrinsic Image Decomposition to generate albedo images and design an albedo supervision loss that uses these albedo maps as training targets for the Albedo Head. We then integrate this loss term into the overall training objective, explicitly exploiting illumination-invariant albedo components to suppress erroneous learning in reflective regions and areas with strong illumination gradients. Experiments on the ScanNetV2 dataset demonstrate that, for the lightweight backbone Lite-Mono, our method achieves an average reduction of 18.5% over the four standard depth error metrics and consistently improves accuracy metrics, without increasing the number of parameters and FLOPs at inference time. Full article

This study develops a voxel-based leaf area estimation framework and validates it using a three-year multi-temporal dataset (2022–2024) of pergola-trained grapevines. The workflow integrates 2D image analysis, ExGR-based leaf segmentation, and 3D reconstruction using Structure-from-Motion (SfM). Multi-angle canopy images were collected repeatedly during the growing seasons, and destructive leaf sampling was conducted to quantify true leaf area across multiple vines and years. After removing non-leaf structures with ExGR filtering, the point clouds were voxelized at a 1 cm³ resolution to derive structural occupancy metrics. Voxel-based leaf area showed strong within-vine correlations with destructively measured values (R² = 0.77–0.95), while cross-vine variability was influenced by canopy complexity, illumination, and point-cloud density. In contrast, optical LAI tools (DHP and LAI–2000) exhibited negligible correspondence with true leaf area due to multilayer occlusion and lateral light contamination typical of pergola systems. This expanded, multi-year analysis demonstrates that voxel occupancy provides a robust and scalable indicator of canopy structural density and leaf area, offering a practical foundation for remote-sensing-based phenotyping, yield estimation, and data-driven management in perennial fruit crops. Full article

Image-guided surgical procedures demand tracking systems that combine high accuracy, low latency, and minimal footprint to ensure safe and precise navigation in the operating room. To address these requirements, we developed a monocular optical tracking system based on a single near-infrared camera with directional illumination and compact retroreflective markers designed for short-range measurement. Small dodecahedral markers carrying fiducial patterns on each face were fabricated to enable robust detection in confined and variably illuminated surgical environments. Their non-metallic construction ensures compatibility with CT and MRI, and they can be sterilized using standard autoclave procedures. Multiple fiducial families, detection strategies, and optical hardware configurations were systematically assessed to optimize accuracy, depth of field, and latency. Among the evaluated options, the ArUco MIP_36h12 family provided the best overall performance, yielding a translational error of 0.44 ± 0.20 mm and a rotational error of 0.35 ± 0.16° across a working distance of 30–70 cm, based on static position estimates, with a total system latency of 32 ± 8 ms. These results indicate that the proposed system offers a compact, versatile, and precise solution suitable for high-accuracy navigated and image-guided surgery. Full article

This work presents a unified benchmarking framework for evaluating classical machine-learning–based heart-rate estimation from remote photoplethysmography (rPPG) across both RGB and near-infrared (NIR) modalities. Despite extensive research on algorithmic rPPG methods, their relative robustness across datasets, illumination conditions, and sensor types remains inconsistently reported. To address this gap, we standardize ROI extraction, signal preprocessing, rPPG computation, handcrafted feature generation, and label formation across four publicly available datasets: UBFC-rPPG Part 1, UBFC-rPPG Part 2, VicarPPG-2, and IMVIA-NIR. We benchmark five rPPG extraction methods (Green, POS, CHROM, PBV, PCA/ICA) combined with four classical regressors using MAE, RMSE, and $R^2$, complemented by permutation feature importance for interpretability. Results show that CHROM is consistently the most reliable algorithm across all RGB datasets, providing the lowest error and highest stability, particularly when paired with tree-based models. For NIR recordings, PCA with spatial patch decomposition substantially outperforms ICA, highlighting the importance of spatial redundancy when color cues are absent. While handcrafted features and classical regressors offer interpretable baselines, their generalization is limited by small-sample datasets and the absence of temporal modeling. The proposed pipeline establishes robust cross-dataset baselines and offers a standardized foundation for future deep-learning architectures, hybrid algorithmic–learned models, and multimodal sensor-fusion approaches in remote physiological monitoring. Full article

To meet the growing demand for automated blueberry harvesting in smart agriculture, this study proposes an improved lightweight detection network, termed M-YOLOv11n, for fast and accurate blueberry maturity detection in complex natural environments. The proposed model enhances feature representation through an improved lightweight multi-scale design, enabling more effective extraction of fruit features under complex orchard conditions. In addition, attention-based feature refinement is incorporated to emphasize discriminative ripeness-related cues while suppressing background interference. These design choices improve robustness to scale variation and occlusion, addressing the limitations of conventional lightweight detectors in detecting small and partially occluded fruits. By incorporating MsBlock and the attention mechanism, M-YOLOv11n achieves improved detection accuracy without significantly increasing computational cost. Experimental results demonstrate that the proposed model attains 97.0% mAP50 on the validation set and maintains robust performance under challenging conditions such as occlusion and varying illumination, achieving 96.5% mAP50. With an inference speed of 176.6 FPS, the model satisfies both accuracy and real-time requirements for blueberry maturity detection. Compared with YOLOv11n, M-YOLOv11n increases the parameter count only marginally from 2.60 M to 2.61 M, while maintaining high inference efficiency. These results indicate that the proposed method is suitable for real-time deployment on embedded vision systems in smart agricultural harvesting robots and supports early yield estimation in complex field environments. Full article

Marine robots operating in low illumination and turbid waters require reliable measurement and control for surveying, inspection, and monitoring. This paper present a sensor-centric visual–inertial simultaneous localization and mapping (SLAM) pipeline that combines low-light enhancement, learned feature matching, and NeRF-based dense reconstruction to provide stable navigation states. A lightweight encoder–decoder with global attention improves signal-to-noise ratio and contrast while preserving feature geometry. SuperPoint and LightGlue deliver robust correspondences under severe visual degradation. Visual and inertial data are tightly fused through IMU pre-integration and nonlinear optimization, producing steady pose estimates that sustain downstream guidance and trajectory planning. An accelerated NeRF converts monocular sequences into dense, photorealistic reconstructions that complement sparse SLAM maps and support survey-grade measurement products. Experiments on AQUALOC sequences demonstrate improved localization stability and higher-fidelity reconstructions at competitive runtime, showing robustness to low illumination and turbidity. The results indicate an effective engineering pathway that integrates underwater image enhancement, multi-sensor fusion, and neural scene representations to improve navigation reliability and mission effectiveness in realistic marine environments. Full article

Background subtraction is a core preprocessing step for video analytics, enabling downstream tasks such as detection, tracking, and scene understanding in applications ranging from surveillance to transportation. However, real-time deployment remains challenging when illumination changes, shadows, and dynamic backgrounds produce heavy-tailed pixel variations that are difficult to capture with simple Gaussian assumptions. In this work, we propose a fully parallel GPU implementation of a per-pixel background model that represents temporal pixel deviations with lognormal distributions. During a short training phase, a circular buffer of n frames (as small as $n=3$) is used to estimate, for every pixel, robust log-domain parameters $(\mu ,\sigma )$. During testing, each incoming frame is compared against a robust reference (per-pixel median), and a lognormal cumulative density function yields a probabilistic foreground score that is thresholded to produce a binary mask. We evaluate the method on multiple videos under varying illumination and motion conditions and compare qualitatively with widely used mixture of Gaussians baselines (MOG and MOG2). Our method achieves, on average, 87 fps with a buffer size of 10, and reaches about 188 fps with a buffer size of 3, on an NVIDIA 3080 Ti. Finally, we discuss the accuracy–latency trade-off with larger buffers. Full article

Unmanned Aerial Vehicles (UAVs) have become essential instruments for precision agriculture and forest monitoring, offering rapid, high-resolution data collection over wide areas. This review synthesizes global advances (2015–2024) in UAV-derived vegetation indices (VIs), combining bibliometric and content analyses of 472 peer-reviewed publications. The study identifies key research trends, dominant indices, and technical progress achieved through RGB, multispectral, hyperspectral, and thermal sensors. Results show an exponential growth of scientific output, led by China, the USA, and Europe, with NDVI, NDRE, and GNDVI remaining the most widely applied indices. New indices such as GSI, RBI, and MVI demonstrate enhanced sensitivity for stress and disease detection in both crops and forests. UAV-based monitoring has proven effective for yield prediction, water-stress evaluation, pest identification, and biomass estimation. Despite significant advances, challenges persist regarding illumination correction, soil background influence, and limited forestry applications. The paper concludes that UAV-derived vegetation indices—when integrated with machine learning and multi-sensor data—represent a transformative approach for the sustainable management of agricultural and forest ecosystems. Full article

Stereo cameras, also known as depth cameras or RGB-D cameras, are increasingly employed in a large variety of machinery for obstacle detection purposes and navigation planning. This also represents an opportunity in agricultural machinery for safety purposes to detect the presence of workers on foot and avoid collisions. However, their outdoor performance at medium and long range under operational light conditions remains weakly quantified: the authors then fit a field protocol and a model to characterize the pipeline of stereo cameras, taking the Intel RealSense D455 as benchmark, across various distances from 4 m to 16 m in realistic farm settings. Tests have been conducted using a 1 square meter planar target in outdoor environments, under diverse illumination conditions and with the panel being located at 0°, 10°, 20° and 35° from the center of the camera’s field of view (FoV). Built-in presets were also adjusted during tests, to generate a total of 128 samples. The authors then fit disparity surfaces to predict and correct systematic bias as a function of distance and radial FoV position, allowing them to compute mean depth and estimate a model of systematic error that takes depth bias as a function of distance, light conditions and FoV position. The results showed that the model can predict depth errors achieving a good degree of precision in every tested scenario (RMSE: 0.46–0.64 m, MAE: 0.40–0.51 m), enabling the possibility of replication and benchmarking on other sensors and field contexts while supporting safety-critical perception systems in agriculture. Full article

Remote photoplethysmography (rPPG) enables non-contact heart rate estimation but remains highly sensitive to illumination variation and dataset-dependent factors. This study proposes CHROM-Y, a robust 2D feature representation that combines chrominance (Ω, Φ) with luminance (Y) to improve physiological signal extraction under varying lighting conditions. The proposed features were evaluated using U-Net, ResNet-18, and VGG16 for heart rate estimation and waveform reconstruction on the UBFC-rPPG and BhRPPG datasets. On UBFC-rPPG, U-Net with CHROM-Y achieved the best performance with a Peak MAE of 3.62 bpm and RMSE of 6.67 bpm, while ablation experiments confirmed the importance of the Y-channel, showing degradation of up to 41.14% in MAE when removed. Although waveform reconstruction demonstrated low Pearson correlation, dominant frequency preservation enabled reliable frequency-based HR estimation. Cross-dataset evaluation revealed reduced generalization (MAE up to 13.33 bpm and RMSE up to 22.80 bpm), highlighting sensitivity to domain shifts. However, fine-tuning U-Net on BhRPPG produced consistent improvements across low, medium, and high illumination levels, with performance gains of 11.18–29.47% in MAE and 12.48–27.94% in RMSE, indicating improved adaptability to illumination variations. Full article