Search Results (45)

Traffic sign segmentation is a fundamental component of intelligent transportation systems and autonomous driving, where reliable pixel-level perception is required under challenging real-world conditions such as illumination variations, occlusion, scale diversity, and complex urban backgrounds. In this work, we propose Residual–Recurrent Kolmogorov–Arnold Network U-Net (R2KAN-U-Net), where “R2” denotes the integration of residual convolutional learning and recurrent KAN-based feature refinement. The proposed architecture combines residual U-Net feature extraction, multi-scale KAN fusion, and recurrent KAN refinement to improve pixel-level traffic sign segmentation under challenging road-scene conditions. The proposed framework integrates three complementary components: (1) residual convolutional blocks for stable feature propagation; (2) a multi-scale KAN fusion bottleneck for capturing contextual information at different receptive fields; and (3) recurrent KAN refinement modules for iterative enhancement of discriminative features. Unlike conventional convolutional architectures, the proposed KAN-based formulation replaces linear transformations with learnable univariate functions, enabling adaptive nonlinear feature modeling. We conduct extensive experiments on a custom dataset containing 9300 annotated urban traffic scene images, as well as on the ADE20K and Cityscapes benchmarks. On the custom dataset, the proposed R2KAN-U-Net achieved a Dice coefficient of 0.92 and an IoU score of 0.89, providing a strong accuracy–efficiency trade-off for traffic-sign foreground segmentation. It achieves competitive segmentation accuracy compared with recent CNN-, transformer-, and state-space-based segmentation models while using fewer parameters and lower computational cost. Additional low-light experiments demonstrate improved segmentation stability, with R2KAN-U-Net achieving the highest low-light Dice score of 0.88 and a competitive low-light IoU of 0.79. Furthermore, the proposed architecture maintains competitive computational efficiency with only 24 M parameters, 44.8 G FLOPs, and near-real-time inference at 13 ms per image. The experimental results demonstrate that integrating KAN-based function-space learning with residual and multi-scale feature refinement provides an effective and computationally efficient solution for robust traffic sign segmentation in complex driving environments. Full article

Autonomous ground vehicle navigation in dynamic real-world environments demands path planning systems that simultaneously accommodate real-time environmental hazards and diverse user-defined objectives requirements that classical algorithms, with their static, single-objective cost functions, cannot fulfil. This paper presents the Semantic Personalised Path Planning (SPPP) system, centred on a novel Semantic Personalised Cost (SPC) algorithm that augments the A* search framework with a dynamically computed personalised cost term. The SPC function integrates eight real-time semantic obstacle categories including traffic congestion, weather severity, road surface conditions, and construction activity with eight user-defined preference dimensions spanning safety, travel time, emergency response, comfort, and battery efficiency. An adaptive scaling mechanism amplifies obstacle penalties near the goal, and a gradient-based weight evolution rule refines preference weights iteratively over successive route segments. The user-defined preference activation directly personalises the routing objective to individual operational needs, with the gradient-based evolution further refining preference alignment over successive route segments. Experiments were conducted in two phases: 500 randomised obstacle configurations on a controlled $8\times 8$ grid, and a real 847-node road graph extracted from OpenStreetMap around SRM Institute of Science and Technology, Kattankulathur, representing a single 1.4 km urban corridor, with obstacle scores derived from live Mapbox Traffic and OpenWeatherMap application programming interface data. Under the full emergency preference scenario, SPPP achieves 94.3% obstacle avoidance versus 31.7% for the Euclidean distance threshold A* baseline, a difference statistically significant at $p < 0.001$ under the Wilcoxon signed-rank test with Cohen’s d ≈ 18.9. Real-world computation time of 1.91 ms on a standard laptop and 3.76 ms on a Raspberry Pi 4 confirms deployability on embedded autonomous vehicle hardware. Full article

To address the challenges of missing fine-grained objects, blurred boundaries, and the suppression of shallow details by deep semantic features during cross-scale fusion in traffic scene semantic segmentation, this paper proposes a discrepancy-guided semantic segmentation method with boundary detail enhancement. First, to improve the semantic completeness of fine-grained regions, a Gated Collaborative Context Module (GCCM) is introduced between the encoder and decoder. By leveraging gating-guided channel selection and multi-scale contextual modeling, GCCM adaptively captures semantic dependencies across different scales. Second, to alleviate boundary ambiguity and detail loss, a Frequency–Edge Guided Enhancement Module (FEGE) is designed in the decoder. This module explicitly models low-frequency structural information and high-frequency edge components via frequency decomposition, and further enhances high-frequency details using the Scharr operator and lightweight convolution, thereby improving the structural representation of object contours and boundary regions. Furthermore, to mitigate the suppression of shallow details during cross-scale feature fusion, a Discrepancy-aware Pixel-Adaptive Gating Fusion module (D-PagFM) is proposed. By jointly modeling feature similarity and local discrepancy, the module adaptively regulates pixel-wise fusion, enhancing detail integration in structurally consistent regions while suppressing misleading fusion in inconsistent regions, thereby improving the robustness of feature fusion and boundary consistency. Experimental results on the Cityscapes and CamVid datasets demonstrate that the proposed method achieves mIoU scores of 80.08% and 82.97%, respectively. Moreover, it shows more significant improvements in boundary-sensitive fine-grained categories such as road boundaries, poles, and traffic signs, indicating its effectiveness and application potential for high-precision semantic segmentation in traffic scenes. Full article

Digital twins (DTs) are increasingly adopted to enhance the monitoring, management, and planning of urban infrastructure. While DT development for buildings is well established, applications to urban road networks remain limited, particularly in integrating heterogeneous geospatial datasets into semantically rich, multi-scale representations. This study presents a methodology for developing a BIM-based DT of urban roads by integrating geospatial data from Mobile Mapping System (MMS) surveys with semantic information from municipal geodatabases. The approach follows a multi-modal (point clouds, imagery, vector data), multi-scale and multi-level framework, where ‘multi-level’ refers to modeling at different scopes—from a city-wide level, offering a generalized representation of the entire road network, to asset-level detail, capturing parametric BIM elements for individual road segments or specific components such as road sign and road marker, lamp posts and traffic light. MMS-derived LiDAR point clouds allow accurate 3D reconstruction of road surfaces, curbs, and ancillary infrastructure, while municipal geodatabases enrich the model with thematic layers including pavement condition, road classification, and street furniture. The resulting DT framework supports multi-scale visualization, asset management, and predictive maintenance. By combining geometric precision with semantic richness, the proposed methodology delivers an interoperable and scalable framework for sustainable urban road management, providing a foundation for AI-ready applications such as automated defect detection, traffic simulation, and predictive maintenance planning. The resulting DT achieved a geometric accuracy of ±3 cm and integrated more than 45 km of urban road network, enabling multi-scale analyses and AI-ready data fusion. Full article

Autonomous vehicles represent a transformative technology in modern transportation, promising enhanced safety, efficiency, and accessibility in mobility systems. This paper presents a comprehensive autonomous vehicle system designed specifically for Vietnam’s traffic conditions, featuring a multi-layered approach to perception, decision-making, and control. The system utilizes dual 2D LiDARs, camera vision, and GPS sensing to navigate complex urban environments. A key contribution is the development of a specialized segmentation model that accurately identifies Vietnam-specific traffic signs, lane markings, road features, and pedestrians. The system implements a hierarchical decision-making architecture, combining long-term planning based on GPS and map data with short-term reactive planning derived from a bird’s-eye view transformation of segmentation and LiDAR data. The control system modulates the speed and steering angle through a validated model that ensures stable vehicle operation across various traffic scenarios. Experimental results demonstrate the system’s effectiveness in real-world conditions, achieving a high accuracy rate in terms of segmentation and detection and an exact response in navigation tasks. The proposed system shows robust performance in Vietnam’s unique traffic environment, addressing challenges such as mixed traffic flow and country-specific road infrastructure. Full article

High-definition vector maps, such as Lanelet2, are critical for autonomous driving systems, enabling precise localization, path planning, and regulatory compliance. However, creating and maintaining these maps traditionally demands labor-intensive manual annotation or resource-heavy automated pipelines. This paper presents an ROS 2-based framework for semi-automatic vector map generation, leveraging Lanelet2 primitives to streamline map creation while balancing automation with human oversight. The framework integrates multi-sensor inputs (LIDAR, GPS/IMU) within ROS 2 to extract and fuse road features such as lanes, traffic signs, and curbs. The pipeline employs modular ROS 2 nodes for tasks including NDT and SLAM-based pose estimation and the semantic segmentation of drivable areas which serve as a basis for Lanelet2 primitives. To promote adoption, the implementation is released as an open source. This work bridges the gap between automated map generation and human expertise, advancing the practical deployment of dynamic vector maps in autonomous systems. Full article

Communications in intelligent transportation systems (ITS) face explosive data growth from applications such as autonomous driving, remote diagnostics, and real-time monitoring, imposing severe challenges on limited spectrum, bandwidth, and latency. Reliable semantic image reconstruction under noisy channel conditions is critical for ITS perception tasks, since noise directly impacts the recognition of both static infrastructure and dynamic obstacles. Unlike traditional approaches that aim to transmit all image data with equal fidelity, effective ITS communication requires prioritizing task-relevant dynamic elements such as vehicles and pedestrians while filtering out largely static background features such as buildings, road signs, and vegetation. To address this, we propose an Offline Knowledge Base and Attention-Driven Semantic Communication (OKBASC) framework for image-based applications in ITS scenarios. The proposed framework performs offline semantic segmentation to build a compact knowledge base of semantic masks, focusing on dynamic task-relevant regions such as vehicles, pedestrians, and traffic signals. At runtime, precomputed masks are adaptively fused with input images via sparse attention to generate semantic-aware representations that selectively preserve essential information while suppressing redundant background. Moreover, we introduce a further Bi-Level Routing Attention (BRA) module that hierarchically refines semantic features through global channel selection and local spatial attention, resulting in improved discriminability and compression efficiency. Experiments on the VOC2012 and nuPlan datasets under varying SNR levels show that OKBASC achieves higher semantic reconstruction quality than baseline methods, both quantitatively via the Structural Similarity Index Metric (SSIM) and qualitatively via visual comparisons. These results highlight the value of OKBASC as a communication-layer enabler that provides reliable perceptual inputs for downstream ITS applications, including cooperative perception, real-time traffic safety, and incident detection. Full article

Assessing the driver’s state in real time is a critical challenge in modern vehicle safety systems, as human factors account for the vast majority of traffic accidents. Heart rate (HR) is a key physiological indicator of the driver’s condition, yet contactless measurements in dynamic in-vehicle environments remain difficult due to motion artifacts, vibrations, and varying operational conditions. This paper presents a novel two-stage method for HR estimation using a commercial 60 GHz frequency-modulated continuous wave (FMCW) radar sensor, specifically designed and validated for in-vehicle applications. In the first stage, coarse HR estimation is performed using the discrete wavelet transform (DWT) and autoregressive (AR) spectral analysis. The second stage refines the estimate using an inverse application of the relevance vector machine (RVM) approach, leveraging a narrowed frequency window derived from Stage 1. Final HR estimates are stabilized through sequential Kalman filtering (SKF) across time segments. The system was implemented using an Infineon BGT60TR13C radar module installed in the sun visor of a passenger vehicle. Extensive data collection was conducted during real-world driving across diverse traffic scenarios. The results demonstrate robust HR estimations with an accuracy comparable to that of commercial wearable devices, validated against a Polar H10 chest strap. This method offers several advantages over prior work, including short measurement windows (5 s), operation under varying lighting and clothing conditions, and validation in realistic driving environments. In this sense, the method contributes to the field of biomimetics by transferring the biological principles of continuous vital sign perception to technical sensorics in the automotive domain. Future work will explore the fusion of sensors with visual methods and potential extension to heart rate variability (HRV) estimations to enhance driver monitoring systems (DMSs) further. Full article

Image segmentation is a foundation for autonomous driving frameworks that empower vehicles to explore and navigate their surrounding environment. It gives a fundamental setting to the dynamic cycles by dividing the image into significant parts like streets, vehicles, walkers, and traffic signs. Precise segmentation ensures safe navigation and the avoidance of collisions, while following the rules of traffic is very critical for seamless operation in self-driving cars. The most recent deep learning-based image segmentation models have demonstrated impressive performance in structured environments, yet they often fall short when applied to the complex and unpredictable conditions encountered in autonomous driving. This study proposes an Adaptive Ensemble Attention (AEA) mechanism within a Generative Adversarial Network architecture to deal with dynamic and complex driving conditions. The AEA integrates the features of self, spatial, and channel attention adaptively and powerfully changes the amount of each contribution as per input and context-oriented relevance. It does this by allowing the discriminator network in GAN to evaluate the segmentation mask created by the generator. This explains the difference between real and fake masks by considering a concatenated pair of an original image and its mask. The adversarial training will prompt the generator, via the discriminator, to mask out the image in such a way that the output aligns with the expected ground truth and is also very realistic. The exchange of information between the generator and discriminator improves the quality of the segmentation. In order to check the accuracy of the proposed method, the three widely used datasets BDD100K, Cityscapes, and KITTI were selected to calculate average IoU, where the value obtained was 89.46%, 89.02%, and 88.13% respectively. These outcomes emphasize the model’s effectiveness and consistency. Overall, it achieved a remarkable accuracy of 98.94% and AUC of 98.4%, indicating strong enhancements compared to the State-of-the-art (SOTA) models. Full article

Autonomous vehicles are a promising solution to traffic congestion, air pollution, accidents, wasted time, and resources. However, remote driver intervention may be necessary in extreme situations to ensure safe roadside parking or complete remote takeover. In these cases, high-quality real-time video streaming is crucial for remote driving. In a preliminary study, we presented a region of interest (ROI) High-Efficiency Video Coding (HEVC) method where the image was segmented into two categories: ROI and background. This involved allocating more bandwidth to the ROI, which yielded an improvement in the visibility of classes essential for driving while transmitting the background at a lower quality. However, migrating the bandwidth to the large ROI portion of the image did not substantially improve the quality of traffic signs and lights. This study proposes a method that categorizes ROIs into three tiers: background, weak ROI, and strong ROI. To evaluate this approach, we utilized a photo-realistic driving scenario database created with the Cognata self-driving car simulation platform. We used semantic segmentation to categorize the compression quality of a Coding Tree Unit (CTU) according to its pixel classes. A background CTU contains only sky, trees, vegetation, or building classes. Essentials for remote driving include classes such as pedestrians, road marks, and cars. Difficult-to-recognize classes, such as traffic signs (especially textual ones) and traffic lights, are categorized as a strong ROI. We applied thresholds to determine whether the number of pixels in a CTU of a particular category was sufficient to classify it as a strong or weak ROI and then allocated bandwidth accordingly. Our results demonstrate that this multi-category ROI compression method significantly enhances the perceptual quality of traffic signs (especially textual ones) and traffic lights by up to 5.5 dB compared to a simpler two-category (background/foreground) partition. This improvement in critical areas is achieved by reducing the fidelity of less critical background elements, while the visual quality of other essential driving-related classes (weak ROI) is at least maintained. Full article

The conservation of biodiversity and the balance between ecological and societal needs are critical but often contested global issues. Wildlife-vehicle collision (WVC) on vital infrastructure, especially linear infrastructure, remains a persistent challenge from policy to practice and poses a serious life-threatening implication to humans and other non-human lives. Addressing this issue effectively requires solutions that provide win-win outcomes from both ecological and societal perspectives. This study critically analyzes a decade of roadkill incidents along Nepal’s longest East-West national highway, which passes through a biologically diverse national park in the western Terai Arc Landscape Area (TAL). Findings are drawn from field-based primary data collection of the period 2012–2022, secondary literature review, key informant interviews, and spatial analysis. The study reveals significant variations in roadkill incidence across areas and years. Despite Bardia National Park being larger and having a higher wildlife density, Banke National Park recorded higher roadkill rates. This is attributed to insufficient mitigation measures and law enforcement, more straight highway segments, and the absence of buffer zones between the core park and adjacent forest areas—only a road separates them. Wild boars (Sus scrofa) and spotted deer (Axis axis), the primary prey of Bengal tigers (Panthera tigris tigris), were the most frequently road-killed species. This may contribute to human-tiger conflicts, as observed in the study areas. Seasonal trends showed that reptiles were at higher risk during the wet season and mammals during winter. Hotspots were often located near checkpoints and water bodies, highlighting the need for targeted mitigation efforts such as wildlife crossings and provisioning wildlife requirements such as water, grassland, and shelter away from the regular traffic roads. Roadkill frequency was also influenced by forest cover and time of day, with more incidents occurring at dawn and dusk when most of the herbivores become more active in search of food, shelter, water, and their herds. The findings underscore the importance of road characteristics, animal behavior, and landscape features in roadkill occurrences. Effective mitigation strategies include wildlife crossings, speed limits, warning signs, and public education campaigns. Further research is needed to understand the factors in driving variations between parks and to assess the effectiveness of mitigation measures. Full article

Comprehensive datasets are crucial for developing advanced AI solutions in road infrastructure, yet most existing resources focus narrowly on vehicles or a limited set of object categories. To address this gap, we introduce the Korean Road Infrastructure Dataset (KRID), a large-scale dataset designed for real-world road maintenance and safety applications. Our dataset covers highways, national roads, and local roads in both city and non-city areas, comprising 34 distinct types of road infrastructure—from common elements (e.g., traffic signals, gaze-directed poles) to specialized structures (e.g., tunnels, guardrails). Each instance is annotated with either bounding boxes or polygon segmentation masks under stringent quality control and privacy protocols. To demonstrate the utility of this resource, we conducted object detection and segmentation experiments using YOLO-based models, focusing on guardrail damage detection and traffic sign recognition. Preliminary results confirm its suitability for complex, safety-critical scenarios in intelligent transportation systems. Our main contributions include: (1) a broader range of infrastructure classes than conventional “driving perception” datasets, (2) high-resolution, privacy-compliant annotations across diverse road conditions, and (3) open-access availability through AI Hub and GitHub. By highlighting critical yet often overlooked infrastructure elements, this dataset paves the way for AI-driven maintenance workflows, hazard detection, and further innovations in road safety. Full article

Accurate traffic sign recognition is one of the core technologies of intelligent driving systems, which face multiple challenges such as insufficient light and shadow interference at night. In this paper, we improve the YOLOv5 model for small, fuzzy, and partially occluded traffic sign targets at night and propose a high-precision nighttime traffic sign recognition method, “NTS-YOLO”. The method firstly preprocessed the traffic sign dataset by adopting an unsupervised nighttime image enhancement method to improve the image quality under low-light conditions; secondly, it introduced the Convolutional Block Attention Module (CBAM) attentional mechanism, which focuses on the shape of the traffic sign by weighting the channel and spatial features inside the model and color to improve the perception under complex background and uneven illumination conditions; and finally, the Optimal Transport Assignment (OTA) loss function was adopted to optimize the accuracy of predicting the bounding box and thus improve the performance of the model by comparing the difference between two probability distributions, i.e., minimizing the difference. In order to evaluate the effectiveness of the method, 154 samples of typical traffic signs containing small targets and fuzzy and partially occluded traffic signs with different lighting conditions at nighttime were collected, and the data samples were subjected to the CBAM, OTA, and a combination of the two methods, respectively, and comparative experiments were conducted with the traditional YOLOv5 algorithm. The experimental results showed that “NTS-YOLO” achieved a significant performance improvement in nighttime traffic sign recognition, with a mean average accuracy improvement of 0.95% for the target detection of traffic signs and 0.17% for instance segmentation. Full article

Applications such as autonomous driving demand real-time and high-precision instance segmentation to accurately identify and understand objects in an environment, including pedestrians, vehicles, and traffic signs. Ensuring a balance between accuracy and efficiency in instance segmentation systems is critical for such tasks. Traditional convolutional models face limitations in capturing complex features and global context effectively. To address these challenges, we propose an enhanced QueryInst-based instance segmentation framework. First, we replace the traditional CNN backbone with the DaViT Transformer to extract richer, multi-scale features. Next, we integrate Feature Pyramid Network CARAFE to capture global context and recover missed instances. Finally, we incorporate the Complete IoU (CIoU) loss function to optimize object localization and improve prediction accuracy. Experiments on the Cityscapes and COCO datasets demonstrate that our approach achieves mIoU scores of 46.7% and AP score of 45.5%, representing improvements of 6.1% and 2.6% over the baseline, respectively, outperforming other state-of-the-art methods. Full article

The utilization of robust, pre-trained foundation models enables simple adaptation to specific ongoing tasks. In particular, the recently developed Segment Anything Model (SAM) has demonstrated impressive results in the context of semantic segmentation. Recognizing that data collection is generally time-consuming and costly, this research aims to determine whether the use of these foundation models can reduce the need for training data. To assess the models’ behavior under conditions of reduced training data, five test datasets for semantic segmentation will be utilized. This study will concentrate on traffic sign segmentation to analyze the results in comparison to Mask R-CNN: the field’s leading model. The findings indicate that SAM does not surpass the leading model for this specific task, regardless of the quantity of training data. Nevertheless, a knowledge-distilled student architecture derived from SAM exhibits no reduction in accuracy when trained on data that have been reduced by 95%. Full article