<rss version="2.0" xmlns:atom="https://www.w3.org/2005/Atom">
  <channel>
    <title>Transport Research International Documentation (TRID)</title>
    <link>https://trid.trb.org/</link>
    <atom:link href="https://trid.trb.org/Record/RSS?s=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" rel="self" type="application/rss+xml" />
    <description></description>
    <language>en-us</language>
    <copyright>Copyright © 2026. National Academy of Sciences. All rights reserved.</copyright>
    <docs>http://blogs.law.harvard.edu/tech/rss</docs>
    <managingEditor>tris-trb@nas.edu (Bill McLeod)</managingEditor>
    <webMaster>tris-trb@nas.edu (Bill McLeod)</webMaster>
    <image>
      <title>Transport Research International Documentation (TRID)</title>
      <url>https://trid.trb.org/Images/PageHeader-wTitle.jpg</url>
      <link>https://trid.trb.org/</link>
    </image>
    <item>
      <title>BallastAttN: Occlusion-Robust 3D Railway Ballast Characterization using Data Synthesis and Deep Learning</title>
      <link>https://trid.trb.org/View/2701226</link>
      <description><![CDATA[Accurate characterization of railway ballast is critical for track safety and maintenance; however, traditional field sampling/sieving or two-dimensional images captured are often labor-intensive and limited for a representative analysis. Three-dimensional (3D) point cloud analysis may offer a more comprehensive approach; the dense packing and heavy occlusion of ballast particles restrict image segmentation. This study introduces a novel deep learning pipeline designed for robust 3D railway ballast characterization, BallastAttN. Its core contributions include a comprehensive synthetic training data set from high-fidelity 3D scans of new and degraded ballast particles, an enhanced point cloud segmentation model upgraded with edge-aware voxelization and curriculum learning, and the novel BallastAttN partial point cloud completion model architected to reconstruct complete particle shapes from the highly incomplete views typical of field conditions. The proposed pipeline was comprehensively validated using controlled laboratory experiments with isolated and clustered configurations of real ballast particles in new and degraded conditions. The results show that BallastAttN consistently outperforms the baseline completion framework that utilizes an encoder–decoder architecture mechanism built on attention mechanisms across commonly used size and morphological properties. The performance gap widened substantially in clustered scenarios that are close to the field ballast appearance, demonstrating the model’s enhanced ability to handle occlusion. The predictions were precise in differentiating between new and degraded ballast based on morphological properties, such as 3D sphericity, the Flat and Elongated Ratio, and the Angularity Index. This study establishes a practical framework for automated ballast inspection, for example, with the use of an innovative ballast scanning vehicle developed, paving the way for more efficient and reliable railway ballast maintenance.]]></description>
      <pubDate>Mon, 11 May 2026 12:24:46 GMT</pubDate>
      <guid>https://trid.trb.org/View/2701226</guid>
    </item>
    <item>
      <title>PillarID: Rethinking Backbone Network Designs for Pillar-Based 3D Object Detection in Infrastructure Point Cloud</title>
      <link>https://trid.trb.org/View/2672792</link>
      <description><![CDATA[In recent years, vehicle-centric point cloud 3D object detection has been widely explored and effectively developed. However, due to differences in the placement of sensors, infrastructure-centric point cloud 3D object detection, which is an important component of the Intelligent Transportation System (ITS), has not received sufficient attention as well as effective network architecture design. Based on the difference in perspective of the infrastructure point cloud, We discover that the roadside point cloud is denser and with a higher coverage compared to the vehicle-side in the pillar representation, resulting in a narrowing of the performance difference between dense pillar and sparse pillar backbone networks in roadside scenes. Inspired by this insight, a network based on the dense backbone is proposed, dubbed PillarID. It utilizes Single-stride Cross-stage Dense-backbone (SCD) to obtains efficient computation through channel degradation, split, and cross-stage connection, and benefits from the rich context of the roadside point cloud based on single-stride. Further, Hierarchical Receptive-field Expansion (HRE) are used to address the receptive field constraints of single-stride backbone. Extensive experiments reveal that our PillarID achieves effective designs in terms of architecture and renders the state-of-the-art performance on the popular large-scale roadside benchmark: DAIR-V2X-I and RCooper. The code is available at https://github.com/zhangzhang2024/PillarID]]></description>
      <pubDate>Thu, 07 May 2026 11:02:30 GMT</pubDate>
      <guid>https://trid.trb.org/View/2672792</guid>
    </item>
    <item>
      <title>LLM-guided multi-modal attention spatio-temporal soft Actor-Critic architecture for underwater unmanned vehicle collision avoidance</title>
      <link>https://trid.trb.org/View/2697600</link>
      <description><![CDATA[Autonomous collision avoidance technology is one of the key technologies for the intelligence of UUVs. This paper aims to study the collision avoidance and three-dimensional path planning problems of UUVs in uncertain marine environments. To this end, a multimodal spatiotemporal attention Soft Actor-Critic framework guided by large language models is proposed. Considering that the original SAC algorithm's Actor-Critic network has a simple structure and insufficient feature extraction capability, the Actor-Critic network architecture is reconstructed. Forward-looking multi-beam sonar observations, carrier kinematics, target information, and historical observations are fused, and combined with CNN, LSTM, and Transformer-based multi-head self-attention and spatial attention to effectively capture long-term dependencies and critical spatial features. To address issues such as sparse early experiences and low exploration efficiency in deep reinforcement learning, a pre-guidance mechanism guided by LLM is designed to alleviate experience sparsity and improve the exploration efficiency in the early stages of training. Additionally, a reward function including distance, heading, speed, and collision penalties is redesigned to improve the collision avoidance decision quality of UUVs under both static and dynamic obstacles. Simulation results show that the proposed method can make more effective collision avoidance decisions compared to other RL algorithms in uncertain marine environments, effectively solving the UUV collision avoidance and three-dimensional path planning problems under such conditions.]]></description>
      <pubDate>Thu, 30 Apr 2026 16:39:45 GMT</pubDate>
      <guid>https://trid.trb.org/View/2697600</guid>
    </item>
    <item>
      <title>Point Cloud Processing for Damage Characterization of Steel I-Sections</title>
      <link>https://trid.trb.org/View/2685576</link>
      <description><![CDATA[When structural members experience impact damage, understanding their geometry is a critical step for assessing damage severity and determining how to maintain functionality and structural safety. Traditional methods for characterizing damage severity require hand measurements of displacement at discrete points along the member length. Terrestrial laser scanning (TLS) offers several advantages over manual measurements, such as more comprehensive data, improved accuracy, reduced need for lane closures, and avoiding work at height. However, TLS point cloud data are unstructured and must be processed to extract displacements. Existing methods for point cloud displacement measurements rely on manual techniques, where a point cloud is carefully manipulated to select individual points, but these approaches are impractical for obtaining a global member displacement curve. This paper presents a new semiautomated modular procedure for processing laser scan point cloud data of steel I-sections to obtain bottom-flange displacement curves along the length of a span. Several existing point cloud processing techniques are incorporated, including outlier removal, voxelization, cross-sectional slicing, and random sample consensus (RANSAC) regression. The proposed procedure measures horizontal and vertical displacements at three cross-sectional locations—the left corner, centerline, and right corner of the bottom-flange surface—so that the flange displacement and tilt can be accurately characterized. A person who has been trained can capture hundreds of measurements along a member length in less than an hour of postprocessing time, whereas existing methods would likely require several workdays to produce the same number of measurements. The procedure was developed using TLS data from steel I-girders damaged in overheight vehicle strikes on highway bridges. Laser scan data from a damaged steel-girder highway bridge in Mt. Vernon, Illinois, are used as a case study, and results from three other steel-girder highway bridges are also presented. The measurement procedure has been designed for implementation in structural engineering practice.]]></description>
      <pubDate>Thu, 30 Apr 2026 09:11:45 GMT</pubDate>
      <guid>https://trid.trb.org/View/2685576</guid>
    </item>
    <item>
      <title>Offline Object Tracking With Joint Point Cloud Registration</title>
      <link>https://trid.trb.org/View/2659102</link>
      <description><![CDATA[To meet regulatory and safety standards, a significant amount of ground-truth data is necessary for data-driven validation of perception algorithms. To overcome the high costs and extensive manual efforts required by existing methods, we propose an automatic offline approach to generate ground-truth data for 3D object poses. Our method solely uses lidar point clouds and matches them jointly, making it applicable to unknown rigid object shapes. We integrate soft 3D motion constraints and iteratively refine the kinematic object states, while correcting the point cloud distortion caused by object motion, to limit both global and local registration errors, respectively. To assess the accuracy of our obtained 3D object poses, we model error bounds under ideal conditions based on the sensitivity of the transformation parameters and the accuracy of a single lidar detection. Our experiments using both simulations and real-world sensor data demonstrate the superior performance of our proposed method within the near range compared to DGPS data.]]></description>
      <pubDate>Wed, 29 Apr 2026 09:10:08 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659102</guid>
    </item>
    <item>
      <title>Image-Guided Outdoor LiDAR Perception Quality Assessment for Autonomous Driving</title>
      <link>https://trid.trb.org/View/2659097</link>
      <description><![CDATA[LiDAR is one of the most crucial sensors for autonomous vehicle perception. However, current LiDAR-based point cloud perception algorithms lack comprehensive and rigorous LiDAR quality assessment methods, leading to uncertainty in detection performance. Additionally, existing point cloud quality assessment algorithms are predominantly designed for single-object scenarios. In this paper, we introduce a novel image-guided point cloud quality assessment algorithm for outdoor autonomous driving environments, named the Image-Guided Outdoor Point Cloud Quality Assessment (IGO-PQA) algorithm. Our proposed algorithm comprises two main components. The first component is the IGO-PQA generation algorithm, which leverages point cloud data, corresponding RGB surrounding view images, and agent objects' ground truth annotations to generate an overall quality score for a single-frame LiDAR-based point cloud. The second component is a transformer-based IGO-PQA regression algorithm for no-reference outdoor point cloud quality assessment. This regression algorithm allows for the direct prediction of IGO-PQA scores in an online manner, without requiring image data and object ground truth annotations. We evaluate our proposed algorithm using the nuScenes and Waymo open datasets. The IGO-PQA generation algorithm provides consistent and reasonable perception quality indices. Furthermore, our proposed IGO-PQA regression algorithm achieves a Pearson Linear Correlation Coefficient (PLCC) of 0.86 on the nuScenes dataset and 0.97 on the Waymo dataset.]]></description>
      <pubDate>Wed, 29 Apr 2026 09:10:08 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659097</guid>
    </item>
    <item>
      <title>A Dynamic 3D Multi-Object Tracking Method Based on Spatiotemporal Features</title>
      <link>https://trid.trb.org/View/2659134</link>
      <description><![CDATA[3D multi-object tracking is one of the important research directions in computer vision and holds significant research value in the field of autonomous driving. Relying solely on single image information or point cloud information is insufficient to overcome tracking challenges in complex scenarios. Currently, multimodal fusion 3D tracking methods still face numerous issues in fusion performance, data association, and trajectory management. Therefore, this paper proposes a dynamic 3D multi-object tracking method based on spatiotemporal features. First, a multi-scale spatial feature embedding fusion network is designed to enhance the weight of critical information within different modal features, thereby improving the prominence of target features. Second, a temporal aggregation embedding module is proposed to address the characteristics of point cloud features and fusion features, enhancing feature alignment when target features are integrated into temporal features, resulting in more robust temporal features. Finally, a multi-stage hybrid affinity dynamic association module and an adaptive dynamic trajectory management module are combined to reduce the impact of similar targets on tracking, which improves the model's ability to perceive target positions in dense scenes, and enhances the robustness of target association matching. Experimental results on the KITTI dataset have demonstrated that the proposed method achieves better tracking performance compared to other state-of-the-art methods.]]></description>
      <pubDate>Wed, 29 Apr 2026 09:10:07 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659134</guid>
    </item>
    <item>
      <title>Omni Point Air: LiDAR and Point Cloud Map-Based Place Recognition and Pose Estimation for Advanced Air Mobility in GNSS-Denied Environments</title>
      <link>https://trid.trb.org/View/2659148</link>
      <description><![CDATA[For Advanced Air Mobility (AAM) systems operating in diverse environments, redundant localization techniques are essential to ensure continuous and safe mission execution. In this study, we propose a 3D place recognition and pose estimation method for AAM using a hemispherical light detection and ranging (LiDAR) sensor. The proposed approach includes a feature extraction method that leverages height differences in surrounding objects, a method for generating local and global descriptors from feature distances, and an efficient geometric verification and localization process through correspondence calculation. Additionally, the method incorporates a process to create a virtual descriptor database using a point cloud map, enabling robust localization in unvisited areas. All procedures are handcrafted, and the performance of the proposed method is validated through comparison with state-of-the-art methods using datasets generated in a simulator. The proposed method achieved over 99.16% average precision (AP) and a 99.99% F1 score in loop closure detection. In pose estimation, it achieved a root mean square error (RMSE) of 0.836 meters or less for position and 0.195 degrees or less for heading. Furthermore, a time analysis on both a general PC and an embedded device confirmed the real-time capability of the proposed method, with an average pose estimation time of 21.70 milliseconds on the embedded device, demonstrating its feasibility for real-time localization in low-power environments.]]></description>
      <pubDate>Wed, 29 Apr 2026 09:10:07 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659148</guid>
    </item>
    <item>
      <title>Few-Shot Bridge Component Recognition Using Point Cloud Data Based on Large CV Model</title>
      <link>https://trid.trb.org/View/2640335</link>
      <description><![CDATA[Semantic segmentation of bridge Point Cloud Data (PCD) is an intermediate process required for tasks such as deformation detection and digital twin. However, existing deep learning-based methods have limited performance because of the lack of training samples and the significant size differences of bridge components. To address these issues, this paper presents a novel unsupervised framework for semantic segmentation of bridge PCD based on a large Computer Vision model. The segment anything model with the zero-shot transfer capability is first employed for PCD segmentation, and then a strong feature extractor is used for few-shot PCD classification to obtain semantic labels. A visible point projection method is proposed to bridge the model gap between PCD and images. Experiment results on a real-world bridge data set showed the proposed method achieved outstanding performance on evaluation metrics of overall accuracy (95.31%), mean precision (97.66%), mean recall rate (92.50%), and mean F1 score (94.24%).]]></description>
      <pubDate>Tue, 28 Apr 2026 12:18:37 GMT</pubDate>
      <guid>https://trid.trb.org/View/2640335</guid>
    </item>
    <item>
      <title>Computing in Civil Engineering 2024: Building Information Modeling, Digital Twins, and Simulation and Visualization</title>
      <link>https://trid.trb.org/View/2695138</link>
      <description><![CDATA[This collection contains 83 peer-reviewed papers on building information modeling (BIM), digital twins, and simulation and visualization.  Topics include: innovations in structures; modular and industrialized construction; simulated processes; simulation in construction; visualization innovation; BIM in practice; BIM specialty tools; blockchain in construction; computing in construction management; digital twins concepts; digital twins in action; inference in point clouds; model content generation; point cloud instance segmentation; point cloud processing and application; reality capture; and specialty BIM.  This collection offers a current overview of the state of computing within the civil engineering space for computing science and civil engineering researchers globally.]]></description>
      <pubDate>Sun, 26 Apr 2026 17:37:53 GMT</pubDate>
      <guid>https://trid.trb.org/View/2695138</guid>
    </item>
    <item>
      <title>Robust and optimised sub-component detection using 3D point cloud for shipbuilding sub-assembly automation</title>
      <link>https://trid.trb.org/View/2693638</link>
      <description><![CDATA[This study proposes a 3D object detection system to reliably detect components in complex environments using 3D point cloud data to automate arrangement operations in shipbuilding sub-assembly processes. To achieve this, an improved RANSAC algorithm was developed by integrating a scoring scheme that considers normal vector similarity, inlier distribution, and the presence of regions containing the topmost feature points. Following the initial model estimation, a refitting procedure based on the inlier set was performed to enhance the stability and accuracy of the plane model. Additionally, a priority-based sampling method was applied to the plane estimation process to optimise computational efficiency. A component-selection algorithm was developed that considered the classification ID, position, size, count, and remaining status of the recognised components, and a fully integrated processing framework was constructed to transmit the recognition results in real time for direct use by a gantry robot. Applying the proposed system to a shipyard sub-assembly environment demonstrated stable recognition performance under stacking conditions, with average dimensional and pose errors of 2.4 mm and 0.35°, respectively. The average processing time was 0.99 s, confirming the suitability of the system for real-time automated processes.]]></description>
      <pubDate>Tue, 21 Apr 2026 14:31:07 GMT</pubDate>
      <guid>https://trid.trb.org/View/2693638</guid>
    </item>
    <item>
      <title>Converting structured data to point cloud data: A traffic accident severity prediction model based on sparse 3D convolution</title>
      <link>https://trid.trb.org/View/2686115</link>
      <description><![CDATA[The convolutional neural network (CNN) is currently the most popular model for predicting traffic accident severity. Existing studies typically convert numerical data into single-channel images using dimensionality reduction techniques, which are then used as input for CNN-based prediction models. In this study, the authors propose a novel 3D-CNN model, T2pNet, which maps traffic accident features to a higher dimensional space, converts them to point cloud data, and adjusts and optimizes the coordinates of points in the point cloud based on feature correlation. Compared to the transformation technique employed in 2D-CNN models, this approach reduces the information loss generated during the dimensionality reduction process and introduces a richer spatial feature representation for the traffic accident data. The authors compare the performance of T2pNet against state-of-the-art models. Results demonstrate that T2pNet outperforms the other models, particularly in predicting the severity of serious and fatal accidents, with a substantial lead in F1 score and recall metrics. Furthermore, the authors visualize the convolutional layers of T2pNet using the gradient-weighted class activation mapping technique, which explains the model’s advantages in capturing the correlation and combination relationships among traffic accident features.]]></description>
      <pubDate>Wed, 15 Apr 2026 10:29:30 GMT</pubDate>
      <guid>https://trid.trb.org/View/2686115</guid>
    </item>
    <item>
      <title>Serialized point cloud segmentation with dilated patchwise attention for generating geometric twins of shield metro tunnels</title>
      <link>https://trid.trb.org/View/2686373</link>
      <description><![CDATA[Accurate and efficient semantic segmentation is fundamental to digital twin generation. To address the limitations of existing methods in terms of segmentation efficiency and accuracy, a transformer-based segmentation framework named SerialFormer is developed. SerialFormer incorporates bitwise spatial mapping to compute ordered encodings of point clouds based on space-filling curves, enabling fast point cloud serialization for segmentation. A dilated patchwise attention mechanism is proposed and integrated into the SerialFormer backbone to improve the segmentation performance for small objects. A segment pose estimation method based on cylindrical fitting and template matching is developed to reconstruct the geometric twins of tunnels. The experimental results reveal that SerialFormer achieves an mIoU of 92.1%, outperforming four existing networks while demonstrating superior computational efficiency. The impact of different serialization patterns on segmentation performance is analysed through an ablation study. Compared with field measurements, the reconstructed geometric twin has an average joint offset error of 2.7 mm.]]></description>
      <pubDate>Tue, 14 Apr 2026 10:09:15 GMT</pubDate>
      <guid>https://trid.trb.org/View/2686373</guid>
    </item>
    <item>
      <title>Development and Accuracy Validation of an Image-Based Pavement Roughness Inspection System</title>
      <link>https://trid.trb.org/View/2658632</link>
      <description><![CDATA[The international roughness index (IRI) is typically obtained using an inertial profiler (IP) to measure pavement elevation. However, the IP has several limitations: it can only measure wheelpaths, requires a constant speed, and displays problems with reproducibility. To overcome these constraints, we developed an image-based roughness inspection system (IPRIS), capable of capturing pavement images at 1-m intervals at speeds of up to 100  km/h. Utilizing photogrammetry, IPRIS reconstructs image positions and orientations to generate three-dimensional (3D) point clouds, producing a high-resolution digital surface model (DSM) and orthophoto. This enables flexible profile selection on the orthophoto, allowing IRI calculations directly from DSM elevation data. To assess the accuracy of the IPRIS, we established a reference elevation profile using a high-precision total station and leveling and then compared the IPRIS with a terrestrial light detection and ranging system (TLS). Results showed that the IPRIS achieved IRI Class 1 specification, with an elevation standard deviation of approximately 1.2 mm and an IRI difference of less than 0.200  m/km. By contrast, the TLS exhibited significant point cloud noise, creating a two-fold increase in elevation errors compared to the IPRIS and an IRI difference of 0.419  m/km, meeting only IRI Class 2 specification. Furthermore, repeated IPRIS data collection on a highway yielded consistent results, demonstrating its reliability for high-speed IRI inspections. However, a key limitation of the IPRIS is its reliance on clear pavement images for 3D reconstruction, making it unsuitable for nighttime or rainy conditions.]]></description>
      <pubDate>Mon, 13 Apr 2026 09:40:12 GMT</pubDate>
      <guid>https://trid.trb.org/View/2658632</guid>
    </item>
    <item>
      <title>A point-cloud based deep learning method for predicting hydrodynamic performance</title>
      <link>https://trid.trb.org/View/2687445</link>
      <description><![CDATA[Hydrodynamic performance prediction of marine ships plays a critical role in ship design and optimization. In this work, we propose a point-cloud based hydrodynamic performance prediction framework, termed ShipPointNet. ShipPointNet directly processes the 3D hull surface in the form of point-cloud. By learning spatial geometric features from raw point-cloud data, the proposed method achieves modeling-method independence, allowing it to be applied to hull forms generated via parametric design, CAD modeling, or reverse engineering. This enhances the generality, flexibility, and accuracy of hydrodynamic performance prediction. This study utilizes the hydrodynamic performance of bare-hull resistance as a benchmark to show the viability and competitiveness of the ShipPointNet framework. Specifically, bare-hull resistance data from 2000 ships were collected, and ShipPointNet's performance and advantages were evaluated by comparison with other approaches. Numerous numerical tests validated the suggested method's accuracy and resilience. To further confirm ShipPointNet, towing tank experiments were performed. The results showed a prediction inaccuracy is 2.011% when compared to physical model tests. These results effectively demonstrate the potential of point-cloud based deep learning methods for predicting ship hydrodynamic performance.]]></description>
      <pubDate>Mon, 13 Apr 2026 09:37:44 GMT</pubDate>
      <guid>https://trid.trb.org/View/2687445</guid>
    </item>
  </channel>
</rss>