Search Results (1,445)

Autonomous vehicles depend on heterogeneous sensing systems whose performance varies with range, illumination, weather, object material, traffic geometry, contamination, calibration quality, and cyber-physical interference. This PRISMA 2020 and PRISMA-S systematic review synthesized 65 peer-reviewed primary studies selected from 2143 records identified through four databases. After removal of 793 records before screening, 1350 titles and abstracts were screened; 273 full texts were assessed and 208 were excluded with documented reasons. The final evidence base covers cameras, LiDAR, radar, thermal and event cameras, GNSS/IMU localization, calibration, synchronization, multimodal fusion, adverse-weather perception, sensor-health monitoring, and fault-tolerant perception. No modality was universally superior: comparative performance depended on hardware generation, dataset, environmental severity, range, and metric. Direct evidence was strongest for component-level perception and controlled degradation, whereas health-conditioned fusion, ODD restriction, and minimum-risk behavior were supported mainly by partial experimental evidence and safety-oriented synthesis. The review therefore proposes, rather than claims to validate, a reliability-aware architecture that separates sensor health from task confidence, preserves uncertainty and provenance, adapts fusion, and constrains operation when residual evidence is insufficient. The review was retrospectively registered in PROSPERO on 30 July 2026 (CRD420261465869). Full article

The latest generation of 4D imaging radar demonstrates significant potential in autonomous driving environmental perception, leveraging its capability to provide target elevation data and dense point clouds. This paper introduces a complete method for constructing a multimodal 4D imaging radar dataset for three-dimensional traffic scenes. It illustrates the hardware and software configurations of the data-acquisition vehicle. Methods including multi-sensor coordination, parameter calibration, timestamp synchronization and spatial datum synchronization are proposed. And eight typical three-dimensional traffic scenarios are designed, such as rainy weather environments, dense heterogeneous targets, enclosed tunnels, high-speed cut-in of multiple vehicles, multi-layered stereoscopic structures and edge working condition reproduction. In addition, this paper puts forward a frame-by-frame processing method for high-resolution images and point cloud data collected by the high-definition camera-LiDAR-4D imaging radar collaborative system. A large model-based 3D annotation method for multiple types of targets is proposed, generating a spatio-temporal sequence-optimized four-dimensional annotation sequence, and finally constructs a complete and high-quality multimodal 4D imaging radar dataset for three-dimensional traffic scenes. The results show that the constructed dataset enables the synchronization of timestamps and spatial coordinate systems. The large model can achieve high-precision 3D annotation for the four predefined target types. The dataset contains 11,400 frames of data from high-definition cameras, LiDAR, and 4D imaging radar, with 131,642 labels. This study will provide reliable fundamental support for the training and verification of 4D imaging radar perception algorithms, vehicle decision-making and planning in complex scenarios, and multi-sensor fusion technologies. Full article

The three-dimensional documentation of crash-damaged vehicles can support the visual and geometric recording of deformation, but it is unclear how complete mobile reconstruction workflows compare when applied to the same vehicles. This study compared three workflows using a single consumer device, an Apple iPhone 16 Pro Max: reconstruction from photographs, reconstruction from extracted video frames, and direct mobile light detection and ranging (LiDAR) scanning. Three damaged vehicles were documented: a Volkswagen Passat B6 Variant, a Toyota Auris, and a Toyota Yaris. RealityScan was used for reconstruction from photographs and video frames, and Polycam was used for the LiDAR scans. In CloudCompare, all models were cleaned, scaled using the known wheelbase, registered to the LiDAR reference by the Iterative Closest Point algorithm, and compared using cloud-to-mesh distances, with the principal quantitative statistics based on absolute point-to-surface distance magnitudes. Because the mobile LiDAR model served as an internal reference rather than as an independent metrological ground truth, the reported values describe residual post-registration point-to-surface deviations and not absolute geometric accuracy. The principal surface evaluation used exactly 100,000 surface-sampled points per evaluated direction and bidirectional cloud-to-mesh calculations. The standardised results did not show a uniform ordering between reconstruction from photographs and reconstruction from video frames. In the reconstruction-to-LiDAR direction, median absolute distances ranged from 0.03082 to 0.03677 m for the Passat, from 0.02900 to 0.03413 m for the Auris, and from 0.05953 to 0.06168 m for the Yaris. Lower reverse-direction median values and the broader upper-tail distributions observed for the Yaris demonstrated the directional character of the surface comparison. The Yaris showed larger, long-tailed deviations concentrated mainly in the rear and left-lateral damaged regions. However, because each damage configuration was represented by only one vehicle, the observed differences cannot be attributed to damage type alone. The three workflows provided complementary geometric and visual information for crash-damaged vehicle documentation, although model fusion and accident-reconstruction parameters were not evaluated in this study. Because only one acquisition was performed for each vehicle–workflow combination, the findings should be interpreted as an exploratory comparison rather than as an assessment of repeatability, operator variability, or measurement uncertainty. Full article

LiDAR–camera fusion is widely used for 3D perception in intelligent connected vehicles, but a clean camera branch does not necessarily compensate for structured LiDAR corruption. We propose CMS-Attack, a cross-modal search framework in which only the LiDAR point cloud is perturbed while the camera input remains unchanged; here, “cross-modal” denotes that a single-modality LiDAR perturbation propagates through the LiDAR–camera fusion process and disrupts the multimodal detector, rather than simultaneous perturbation of both modalities. The framework has the following two access-dependent routes: the gray-box route contains FB-CMS, which uses camera-BEV, LiDAR-BEV, fused-BEV, and detection-head responses to construct a target-aware prior and prune an over-complete candidate pool, and Adaptive CMS, which substitutes architecture-specific intermediate responses; the decision-only black-box route contains FC-CMS, which refines candidates solely from display-level target states. On the nuScenes validation split, FB-CMS reduced matched target confidence from 0.80 to 0.03 under 140 injected points, corresponding to a 96.1% relative drop and 100% ASR@0.3. The ten-query FC-CMS achieved 58.97% ASR@0.3, compared with 6.17% for random frustum spoofing. On the query-based FUTR3D detector, Adaptive CMS reduced the mean matched-target score from 0.642 to 0.058, corresponding to a 90.97% relative reduction and 93.42% ASR@0.3. Intermediate camera-, LiDAR-, and fused-BEV region energies changed by less than 0.8% despite target suppression, indicating disruption at the fusion-decision stage (defined here as the decoder/detection-head and post-processing path from fused representations to final object predictions) rather than a collapse of BEV feature magnitude. These results show that structured LiDAR fabrication is substantially more disruptive than information removal and that generic outlier filtering incurs a robustness–accuracy tradeoff. Full article

Existing reviews on AI in tea production are either agriculture-generic or limited to isolated tasks. This review thoroughly compares vision technologies (RGB, hyperspectral, near-infrared, thermal, Light Detection and Ranging (LiDAR), Unmanned Aerial Vehicle (UAV)) and establishes a task-oriented algorithm selection framework for the tea industry. For small-sample or near-linear problems, traditional machine learning (ML) (support vector machine (SVM); partial least squares regression (PLSR)) remains effective. For unstructured field tasks, deep learning achieves superior performance: pest detection accuracy exceeds 97%, tea bud detection reaches 96.8% with RGB images, and hyperspectral imaging predicts nitrogen content with R² > 0.90 and tea polyphenols with R² up to 0.925. Algorithm choice further differentiates by task granularity: lightweight convolutional neural networks (CNNs) balance speed and accuracy for edge deployment at 16 fps; You Only Look Once (YOLO) series detectors enable real-time localization on mobile platforms at 93.1% accuracy, 24 ms per target. No single algorithm dominates all tea tasks; selection is a trade-off among accuracy, speed, data availability, and computational constraints. These findings outline a structured analysis of the challenges and pathways for transitioning computer vision (CV) from laboratory research toward field-deployable tools. Full article

Unmanned aerial vehicle (UAV)-based multispectral remote sensing has emerged as a promising tool for monitoring crop physiological status and supporting precision disease management. However, the capacity of multispectral vegetation indices to detect foliar diseases under commercial field conditions remains insufficiently understood, particularly in perennial crops grown in humid tropical environments. This study evaluated the potential of UAV-derived multispectral vegetation indices to assess the physiological response of avocado (Persea americana Mill.) canopies affected by anthracnose caused by Colletotrichum fructicola in Amazonas, Peru. Disease incidence and severity were assessed through field evaluations, while multispectral imagery was acquired using a UAV equipped with a MicaSense RedEdge-MX Dual sensor. Vegetation indices including the Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Red Edge Index (NDRE), Red Edge Chlorophyll Index (CIred-edge), Plant Senescence Reflectance Index (PSRI), and Visible Atmospherically Resistant Index (VARI) were calculated, and their relationships with anthracnose incidence were analyzed using Spearman’s rank correlation. Field observations confirmed a high incidence of foliar anthracnose with a heterogeneous spatial distribution across the orchard. Multispectral imagery successfully characterized spatial variability in the canopy physiological condition, revealing differences in vegetation vigor, chlorophyll-related reflectance, and senescence among experimental blocks. Nevertheless, none of the evaluated vegetation indices showed statistically significant correlations with anthracnose incidence (p > 0.05), indicating that spectral variability primarily reflected the general canopy physiological status rather than a disease-specific spectral response. These findings demonstrate that UAV-derived multispectral vegetation indices are valuable for monitoring spatial variability in the avocado canopy condition but have limited capability for independently detecting anthracnose under humid tropical field conditions. Future studies integrating multi-temporal UAV acquisitions, hyperspectral and thermal imagery, LiDAR, environmental variables, and machine-learning approaches are expected to improve the early detection and spatial prediction of anthracnose in avocado production systems. Full article

Coastal warm fog poses a serious threat to the safety and efficiency of port shipping. However, its artificial dispersal and the effect evaluation of the dispersal remain a worldwide challenge. This paper conducted a field experiment of artificial fog dispersal using an unmanned aerial vehicle (UAV) to spray a new hygroscopic catalyst in Jinshan District, Shanghai, China. A traversing window method was proposed to quantitatively evaluate the dispersal effect of the catalyst on coastal warm fog using the RHI detection mode of a visibility lidar. The experimental results show that the operation period was a typical coastal radiation fog process with a stable meteorological background field. The near-surface wind was weak and the wind direction was constant southeast. The new composite hygroscopic catalyst had a certain effect on coastal warm fog dispersal, and its effect was closely related to the background fog concentration. The operation effect became prominent when the background fog concentration decreased. The maximum visibility improvement reached 456 m, and the peak effect appeared 2–3 min after the end of spraying. The best location for dispersal effect was observed below the operation position and in the downwind direction. The traversing window method proposed in this paper effectively reduces the contingency and spatial representativeness limitations of single-point evaluation. It can locate the optimal area of fog dispersal effect and quantify its intensity. This study provides technical support for the operational application of artificial fog dispersal in coastal area. Full article

The accurate estimation of forest aboveground biomass (AGB) typically relies on extensive field surveys, which are highly time-consuming and cost-prohibitive. While unmanned aerial vehicle (UAV) Light Detection and Ranging (LiDAR) provides ultra-high point densities capable of reliable individual-tree analysis, its limited flight coverage restricts large-scale applications. Conversely, regional airborne laser scanning (ALS) offers broad spatial coverage, but its relatively low point cloud density makes individual-tree level analysis unreliable. To bridge this scale and data gap, this study develops a scale-consistent framework that integrates UAV-LiDAR, three-dimensional simulation, multisource remote sensing, and machine learning for Chinese fir (Cunninghamia lanceolata) plantation AGB estimation. High-density UAV-LiDAR data were first used to construct individual-tree AGB models, and the predicted tree-level biomass was aggregated to generate spatially representative “agent plots” for stand-scale modeling. A three-dimensional (3D) radiative transfer simulation framework was further employed to reproduce airborne LiDAR observations under different point densities, enabling the evaluation of structural information loss caused by LiDAR sparsity. Structural features derived from simulated LiDAR and spectral information from Sentinel-2 imagery were integrated using the Tabular Prior-data Fitted Network (TabPFN). Model reliability was assessed through 10-fold spatial block cross-validation and Monte Carlo simulations, which quantified spatial generalization and uncertainty propagation from individual-tree estimation to stand-level prediction. Feature interpretation using SHapley Additive exPlanations (SHAP) revealed that the LiDAR-derived vertical canopy structure provided the primary constraints for biomass estimation, whereas Sentinel-2 shortwave infrared features supplied complementary information related to canopy conditions. The optimal TabPFN model achieved a stand-level accuracy of R² = 0.88 and RMSE = 9.23 Mg·ha⁻¹ using LiDAR combined with Sentinel-2 data. Uncertainty analysis further demonstrated the robustness of the proposed framework under propagated errors, highlighting its potential for scalable and reliable forest biomass estimation in data-limited subtropical ecosystems. Full article

Protection levels for autonomous vehicle localization are traditionally derived from estimator covariances under Gaussian assumptions. These approaches fail in complex urban environments where sensor anomalies produce heavy-tailed, non-Gaussian error distributions. This paper presents a fundamentally different paradigm that reformulates integrity monitoring as a closed-loop control problem. The method computes an instantaneous error rate from three sources: inertial sensor noise, kinematic drift between filter-based and dead-reckoned displacement, and LiDAR scan-map registration quality weighted by a sensitivity factor. This rate drives a saturation-controlled setpoint dynamics, then an adaptive PID controller with entropy-based gain scheduling produces the final protection level. Asymmetric update laws enforce rapid expansion but cautious contraction of safety bounds. Experiments on three UrbanNavDataset sequences (medium-urban, low-urban, deep-urban) demonstrate that traditional covariance-based methods exhibit high integrity risk, while the proposed framework achieves $0.0\%$ risk in moderate environments and $2.3\%$ under extreme degradation. The resulting protection levels are smooth and well-behaved, compatible with modern motion planners. This control-theoretic approach offers a viable alternative to statistical integrity paradigms in challenging real-world conditions. Full article

Nighttime driving and adverse weather expose persistent weaknesses in autonomous-driving perception pipelines. Low illumination, fog, rain, snow, glare, wet-road reflections, and motion blur degrade camera, LiDAR, radar, and event-camera inputs in modality-specific ways. This review examines robust perception under low-visibility conditions as a pipeline-level problem requiring joint consideration of image enhancement, sensor fusion, detection, and evaluation. Rather than treating enhancement as an isolated restoration task, it analyzes whether recent methods preserve detector-relevant structures, exploit cross-sensor complementarity, and improve downstream 2D and 3D perception. A taxonomy-driven narrative approach compares representative studies along five axes, from input modality and supervision strategy to evaluation protocol and deployment feasibility, supported by a structured verification search of literature published between January 2020 and July 2026, with the search strategy, eligibility criteria, and corpus composition documented. Recent work indicates a shift from image-quality-oriented restoration toward perception-driven optimization, in which enhancement and fusion modules are evaluated by their effect on object detection and 3D perception. Remaining challenges include generalization to compound degradations, cross-sensor misalignment, scene-dependent sensor reliability, latency on in-vehicle edge platforms, and inconsistent benchmark protocols. Future systems should therefore jointly model degradation severity, sensor reliability, downstream task performance, and real-time constraints rather than optimizing restoration, fusion, and detection modules in isolation. Full article

Reliable navigation in GPS-denied environments remains a critical challenge for autonomous vehicles (AVs), particularly in complex indoor and urban settings. GPS-based localization systems often fail under these conditions, highlighting the need for resilient multimodal solutions. In this article, we present a radar-assisted tracking system that integrates LiDAR and inertial measurements within a sensor-fusion architecture to achieve robust navigation. The principal methodological contribution is a unified tracking and prediction framework that combines Bayesian state estimation with learning-based temporal prediction, enabling accurate tracking while continuously forecasting the slave robot’s short-term future state from mapping observations generated by the master robot, with a typical end-to-end perception-to-action latency of 20–60 ms. The communication and prediction forecasting module operates with an update interval below 35 ms, enabling real-time cooperative robotic operation. Sensor data are fused through a pipeline incorporating Gaussian Mixture Models (GMMs) for post-processing, which helps mitigate the limitations associated with individual sensors during edge processing. Moreover, Kalman filtering is employed to mitigate sensor noise and drift, thereby improving state estimation accuracy through trajectory smoothing. The fused spatiotemporal information is subsequently exploited by a Convolutional Recurrent Neural Network (CRNN) coupled with a Nonlinear Autoregressive model with eXogenous Inputs (NARX) to model the robot’s motion dynamics and provide short-horizon state prediction. Through simulations and real-world indoor experiments conducted in GPS-denied environments, we validate the system’s ability to provide accurate and continuous pose estimation with low localization errors. Experimental results show that the proposed framework achieves root-mean-square errors of 0.12 m, 0.15 m, and 0.28 m along the X, Y, and Z axes, respectively, while maintaining sub-meter maximum position deviations throughout the evaluated trajectories. These results confirm that the proposed framework provides reliable localization and predictive state estimation for cooperative robotic navigation in indoor GPS-denied environments. Future work will investigate outdoor validation and extend the framework to additional data-driven decision-making models for future robotic services. Full article

This paper addresses the problem that autonomous bicycles struggle to achieve precise obstacle avoidance and robust dynamic balance at the same time in rugged terrain and narrow constrained spaces. A hybrid hierarchical control architecture that integrates Rapidly-exploring Random Trees (RRT*) with a dual-layer Proximal Policy Optimization (PPO) scheme is proposed, referred to hereafter as the RRT*-2LPPO architecture. In the global planning layer, RRT* combined with cubic B-spline smoothing generates $C^2$-continuous reference trajectories. In the local control layer, decision-making and execution are hierarchically coupled through a dual-layer cooperation scheme. The upper-layer PPO network incorporates LiDAR data and uses the vehicle body pitch angle to enhance rugged terrain perception for heading planning, while the lower-layer PPO network coordinates the momentum wheel and the steering mechanism to maintain vehicle stability. Validation across various challenging scenarios demonstrates that the proposed framework achieves excellent robustness and consistently attains the highest navigation success rates, significantly outperforming traditional control methods and non-hierarchical architectures. Full article

Single-tree diameter at breast height (DBH) and tree height (H) are fundamental parameters for forest inventory, forest structure characterization, and forest carbon stock estimation. However, single-source LiDAR data cannot simultaneously capture complete trunk and canopy structural information, limiting the accuracy of single-tree structural parameter estimation. To address this issue, a Bi-Directional Cross-Attention Fusion Network (BCAF-Net) is proposed to estimate DBH and H separately by integrating ground-based Simultaneous Localization and Mapping LiDAR (SLAM LiDAR) and Unmanned Aerial Vehicle LiDAR (UAV LiDAR) data. The framework employs a dual-branch encoder and a bidirectional cross-attention mechanism to establish cross-view structural relationships between trunk and canopy observations, enabling effective multi-source feature fusion. Experiments conducted at two urban forest sites demonstrated that BCAF-Net achieved the highest estimation accuracy, with RMSE of 0.82 cm for DBH and 0.91 m for H and corresponding R² values of 0.97 and 0.96, respectively. Furthermore, the model maintained stable performance under varying forest structural complexities, cross-site conditions, and tree species. These results demonstrate that cross-view structural interaction effectively exploits complementary information from SLAM LiDAR and UAV LiDAR data, thereby improving single-tree structural parameter estimation in complex forest environments. Full article

Urban low-altitude unmanned aerial vehicle (UAV) navigation combines partial observability, building occlusion, wind disturbance, and continuous control. This study develops and evaluates HW-GA-SAC, a multimodal history-window Soft Actor-Critic (SAC) policy for procedurally generated three-dimensional MuJoCo cities. A Gated Transformer-XL (GTrXL)-inspired gated-attention encoder processes a fixed eight-step navigation history, while a current-frame safety branch supplies vertical clearance, sparse Light Detection and Ranging (LiDAR)-like range sectors, and handcrafted safety cues directly to the actor and critic. The policy uses obstacle-related observations and reward shaping to support collision avoidance; it does not include constrained policy optimization or a separate runtime safety filter. In a seven-method comparison using five training seeds and five evaluation layouts, HW-GA-SAC achieved a 96% ± 3% success rate, 207 ± 16 average return, and 3% ± 4% timeout rate. Feedforward SAC achieved 92% ± 11% success and a 7% ± 10% timeout rate, but its successful paths were more direct. Five-seed learning curves, city-split evaluation, wind sensitivity, sensing perturbations, inference profiling, and ablation studies further characterize the method. Within this simulation protocol, HW-GA-SAC provides the strongest completion-oriented performance, with a measurable trade-off between task completion and path directness. Full article

Quarry waste piles are heterogeneous engineered embankments that are susceptible to slope instability, yet early detection of pre-failure surface changes remains challenging due to complex surface conditions and measurement uncertainty. This study presents an integrated remote sensing-based framework for monitoring quarry waste pile instability by combining multi-temporal change detection with scale-dependent surface roughness analysis. The original contribution of the proposed framework lies in linking displacement-based change detection with multi-scale characterization of surface roughness, enabling both observed surface movement and topographical conditions associated with developing instability to be evaluated within a unified monitoring approach. Multi-epoch Unmanned Aerial Vehicle (UAV)-mounted Light Detection and Ranging (LiDAR) and photogrammetric point clouds were acquired before and after documented failure events at an active quarry site at active quarry sites located northeast of Montreal, Quebec, Canada. The regional climatic conditions, characterized by seasonal freeze–thaw cycles, rapid snowmelt, and periods of heavy rainfall, can promote water infiltration and elevated pore-water pressures, thereby increasing the susceptibility of these heterogeneous waste piles to slope instability. A standardized workflow was implemented, including precision alignment using a Recursive Iterative Closest Point (R-ICP) registration strategy, vegetation filtering with a multiscale CANUPO classifier, and uncertainty quantification through a Level of Detection (LoD) analysis. The resulting LoD thresholds were 10–15 cm for LiDAR-to-LiDAR comparisons and 34–36 cm for mixed-sensor datasets. Multi-scale roughness analysis revealed that zones which later experienced instability exhibited consistently higher and more heterogeneous roughness than adjacent stable areas within a well-defined linear scale range. A roughness-based A/D indicator enabled objective delineation of hazardous zones prior to failure. Post-failure monitoring showed surface smoothing following major displacement, followed by renewed roughness increases associated with secondary movements. These results demonstrate that scale-dependent roughness provides complementary information to displacement-based change detection, enabling potentially unstable areas to be identified and prioritized before substantial displacement becomes evident. The integrated framework can assist quarry managers in targeting field inspections and monitoring efforts toward higher-risk areas and support earlier preventive actions to reduce slope-failure risk. Full article