Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Jesús Balado has a total of 156 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Channel-separated validation of the human-review layer in automated rail inspection under acquisition shift

    by Saeed MohammadKhah, Joaquim Tinoco, Jesús Balado, José C. Matos

    Artificial-intelligence-based visual inspection systems increasingly use auxiliary confidence and evidence signals to rank predictions for limited human review. In safety-relevant engineering inspection, however, these review layers are commonly validated end-to-end under image degradation, even though the same degradation simultaneously changes both the evidence signal and the classifier error population. This study introduces a channel-separated validation framework for trustworthy AI-based inspection, using railway surface-defect assessment as the engineering application. A 2×2 audit independently perturbs the task and evidence pathways and evaluates the incremental utility of the evidence layer against a matched softmax-derived baseline under group-disjoint nested cross-validation. On 932 rail-head images, severe sensor noise caused a complete conclusion reversal: with the classifier error population held fixed, the clean-calibrated engineering evidence fell 0.326 AUROC below the baseline, whereas conventional end-to-end evaluation placed the same evidence 0.397 above it. The reversal occurred in all 25 repeated folds and was independently reproduced with a self-supervised DINOv2 Vision Transformer (-0.305 versus +0.532). The effect was corruption-dependent and was not reproduced under motion blur. Rare-class removal and multiclass decision-geometry analyses showed that sensor-noise-induced reshaping of the error population largely generated the apparent end-to-end benefit, while a prospectively locked DINOv2 analysis supported the proposed geometric boundary condition. A corrected public RSDD experiment provided an external boundary case in which no comparable reversal occurred. These results show that end-to-end robustness testing alone can mischaracterize the reliability of human-review evidence layers and motivate channel-separated auditing before deployment of AI inspection systems.

  • Open Access

    Circular Assessment of Arctic Timber Façades: Integrating Solar Analysis, Active Hyperspectral Sensing, and Ground-Penetrating Radar

    by Alejandro Fernández-Copóns, Ourania Patsia, Carolina Santos-Silva, Francisco Senna-Vieira, Alex Novo, Pedro Arias, Jesús Balado

  • Open Access

    Direct 3D extreme rainfall risk mapping from urban aerial LiDAR point clouds

    by Jesús Balado, Filip Biljecki

    Flooding is one of the most frequent and damaging natural hazards, increasingly intensified by rapid urbanization and extreme precipitation events. Accurate modelling of water accumulation and runoff in dense urban environments remains challenging due to complex geometries and the limitations of conventional surface-based representations. Most approaches require extensive LiDAR preprocessing before hydrological analysis. This study introduces a methodology for direct flood risk analysis on Aerial Laser Scanning (ALS) point clouds, avoiding intermediate surface reconstruction steps. The proposed approach enriches points with indicators of water accumulation and surface runoff potential. Water accumulation is estimated through elevation differences within a 100-m local neighbourhood to detect topographic depressions, while runoff potential is derived from three geometric descriptors (verticality, planarity, and scattering) that describe the capacity of urban structures to channel or obstruct water flow. The method was evaluated using open ALS datasets from three cities with contrasting urban morphologies: Vigo (Spain), Rotterdam (The Netherlands), and Hong Kong (China). Results demonstrate the capacity of the enriched point clouds to identify flood-prone urban features such as low-lying harbour areas, semi-tunnels, and urban canyons, as well as runoff pathways including sloped roofs, stairways, ramps, and road surfaces. The resulting point-level risk representation shows strong agreement with historical flood events and traditional multi-criteria assessments. By using LiDAR point clouds as an analytical medium, the framework enables intuitive 3D flood risk mapping while preserving their geometric richness.

  • Open Access

    Multimodal Large Language Models to road inventory with non-photorealistic Point Cloud visualization

    by Horia Ameen, Mario Soilán, Henrique Lorenzo, Jesús Balado

    Accurate road inventories are crucial for maintenance, safety, and resource allocation, with automation improving efficiency but often lacking user-friendly human-machine interaction. This paper evaluates how non-photorealistic rendering of 3D point clouds impacts Multimodal Large Language Models (MLLMs) interpretation for road inventory, testing three methods on real road data in Santarém (Portugal). From 3D point clouds coloured with RGB information, non-photorealistic techniques are implemented and compared: Ambient Occlusion (AO), Eye-Dome Lighting (EDL) and Multi Feature-Rich Synthetic Color (MFRSC). Several state-of-the-art MLLMs are also tested: GPT5, Gemini2.5-Pro, Gemini2.5-Flash, CogVLM2, MiniCPM-V, Llama4-scout-17b, Mistral-Small3.2, Qwen 2.5vl and Gemma3. The results indicate that non-photorealistic techniques do not hinder the identification of road elements by MLLMs, indicating their potential for 3D point cloud classification tasks even when true RGB colour is not available. Furthermore, the overall performance metrics, with F-scores over 80% for proprietary, state-of-the-art models (GPT5, Sonnet 4.5 and Gemini) show that 2D captures of 3D point clouds can be a suitable data source for zero-shot object classification. Rather than proposing new algorithms, this work contributes an empirical evaluation of how non-photorealistic point-cloud visualizations affect VLM-based road inventory interpretation.

  • Open Access

    Cross-Sensor Robustness sand Spatial Generalization for 3D Railway Point Cloud Semantic Segmentation

    by Arshia Ghasemlou, Mario Soilán, Jesús Balado, Belén Riveiro

    Accurate semantic segmentation of 3D railway point clouds is essential for enabling automated inspection and asset management. Although recent deep learning (DL) models achieve strong performance on large benchmark datasets, their ability to generalize to point clouds captured with different sensors and in different spatial environments remains insufficiently explored. This study investigates the cross-sensor robustness and spatial generalization of state-of-the-art DL architectures for 3D semantic segmentation in railway scenarios. Three advanced models, Point Transformer V3, MinkUNet, and Swin3D, were trained on the SemanticRail3D dataset and evaluated on a newly acquired railway section scanned using three heterogeneous LiDAR systems: a terrestrial laser scanner (Faro Focus S150+), and two handheld mobile mapping devices (CHCNAV RS10 and GeoSLAM ZEB Go). The test area was manually annotated to provide high-quality ground truth for quantitative assessment. Results show substantial performance variations across sensors, driven by differences in point density, noise levels, and scanning geometry. Domain-shift effects were evaluated directly from the model prediction outputs, including per-class IoU differences, uncertainty patterns, and cross-model agreement across sensors. To improve the robustness, an ensemble fusion strategy is evaluated to mitigate cross-sensor variability. The findings highlight the challenges of deploying DL models in real-world railway environments and provide insights for improving sensor-agnostic segmentation pipelines.

  • Implementation of source-grounded Generative AI as support for flipped learning

    by Mario Soilán, Jesús Balado

    This paper reports the implementation of a Generative AI (GenAI)-augmented flipped classroom in a fourth-year Topography and Construction course (bachelor’s degree in Mechanical Engineering). Two student teams prepared and taught theoretical topics using source-grounded GenAI tools to synthesize instructor-curated sources and produce a report and presentation. After each peer-teaching session, students completed a 15-item quiz generated from the presenters’ materials. Average scores were high and similar for presenters and the audience. A post-activity survey showed predominant use of NotebookLM as a GenAI tool and positive perceived usefulness for learning and text preparation, alongside moderate editing and only moderate trust without verification. The results support the feasibility of pairing source-bounded GenAI with peer teaching and immediate formative assessment, while reinforcing the need to teach verification and critical evaluation.

  • Open Access

    A vision-language based image measurement framework for quantitative assessment of railway surface defects

    by Saeed Mohammadkhah, Joaquim Tinoco, Jesús Balado-Frías, José Campos e Matos

  • Open Access

    A Scoping Review of LiDAR Solutions for Urban Safety of Vulnerable Road Users

    by Juan Castrillo, Mario Soilán, Natalia Caparrini, Jesús Balado

    Vulnerable Road Users (VRUs) are involved in a significant proportion of traffic fatalities, and they are highly exposed to severe injuries in urban traffic environments. For detecting and tracking VRUs, LiDAR technology offers precise 3D perception capabilities, overcoming challenges posed by their small size, dynamic behavior, and frequent presence in occluded or congested areas. This work aims to conduct a scoping review of LiDAR-based solutions for preventing and reducing accidents involving VRUs, synthesizing current methodologies, evaluating detection and tracking approaches, and identifying strategies to improve urban safety through data-driven interventions. An analysis of 49 publications indicates that effective monitoring of VRUs depends on a strategic balance between technological performance and practical limitations, such as system costs, calibration complexity, and hardware constraints. Privacy-preserving techniques, such as anonymization and LiDAR-based sensing, are essential to enable ethically responsible large-scale data collection.


  • Open Access

    3D As‐Built Environments in Extended Reality Applications: A Systematic Review

    by Jesús Balado, Yu Feng, Zhouyan Qiu, Weixiao Gao, Arttu Julin

    Accurate integration and navigation of real‐world 3D spaces are fundamental for next‐generation Extended Reality (XR) systems, enhancing immersion, utility, and fidelity. This paper systematically reviews XR workflows using PRISMA guidelines, focusing on 3D data acquisition, modeling, visualization, and user interaction, based on 96 journal publications. Data collection for XR relies on photogrammetry, RGB‐D cameras, and LiDAR, often enhanced by multi‐sensor fusion, although real‐time transmission and semantic alignment remain challenging. XR pipelines are dominated by Building Information Modeling (BIM) software and game engines, frequently integrating Computer‐Aided Design (CAD) models and 3D scanned data. Visualization varies from photorealistic renderings to schematic representations, with Virtual Reality headsets favored for training and Augmented Reality devices applied in inspection and navigation. Interaction paradigms encompass controllers, gestures, gaze, voice, and haptics, with increasing reliance on Artificial Intelligence for multimodal fusion and processing. Despite progress, key challenges persist, including bandwidth limitations, manual 3D modeling, hybrid data management, interoperability issues, and scarcity of open‐source solutions. Additional identified barriers involve balancing visual quality with performance in specific contexts, limited accuracy of non‐invasive Brain‐Computer Interfaces, and restricted market acceptance due to high costs. Overall, XR adoption remains constrained by technical, usability, and accessibility gaps.