From field measurements and laboratory experiments to engineering models and AI. Explore the work behind our infrastructure.
Explore all publications ↗
A pavement-distress detector informed by an audit of road-survey imagery. LearnableContrast adjusts local contrast, while EdgeSPD uses edge information during downsampling. The study compares five model scales with matched baselines; its findings are reported in an arXiv preprint.
Pawarotorn Chaipetch & coauthors
Vehicle-based pavement-survey images
YOLO26-RD · LearnableContrast + EdgeSPD
Locations and classes of pavement distress
Preprint. Compares five model scales against matched YOLO26 and YOLOv12 baselines; results depend on the reported dataset and evaluation setup.
Repository linked by the paper’s authors.
The authors publish code, configurations, checkpoints and audit scripts. The complete commercial survey-image dataset cannot be redistributed.
Publication source ↗A lightweight pavement-damage detector based on YOLOv9 with partial attention. The study evaluates accuracy and inference speed using road images from multiple countries and an expanded set of damage classes. The journal article and its earlier arXiv version are grouped as one work.
Pawarotorn Chaipetch & coauthors
Road images from multiple countries
YOLOv9 + partial attention
Pavement-damage classes and bounding boxes
Published journal article with an earlier arXiv version. Evaluates detection accuracy, inference speed and the effect of the attention block.
Repository linked by the paper’s authors.
The authors’ repository provides training and inference code and links to model weights. See its GPL-3.0 license and setup instructions.
Publication source ↗Combines YOLO9tr damage detection with calibrated monocular depth estimation to assess pavement damage using a vehicle-mounted camera. The conference manuscript investigates both damage classification and dimensional measurement, extending the detector toward road-maintenance assessment.
Pawarotorn Chaipetch & coauthors
Images from a vehicle-mounted single-lens camera
YOLO9tr + calibrated monocular depth estimation
Damage classification and dimensional estimates
Author-uploaded conference manuscript. Reports calibrated measurements over a 3.5–7.0 m range; performance is specific to the described experiment.
The manuscript is available from the author’s ResearchGate upload. A dedicated repository for the complete detection-and-measurement pipeline has not been verified.
Publication source ↗
A road-asset detector combines a pretrained Vision Transformer with YOLOX and a feature-pyramid decoder. The study evaluates detection on Thai highway imagery, including small roadside objects such as kilometer stones.
Weerachai Wongweeranimit · Mr. Sittinun Thongbai & coauthors
Thai highway images
Vision Transformer + YOLOX + FPN
Road-asset detections
Evaluated against YOLOv5 variants on the Thai highway corpus.
Base detector by the YOLOX authors; this is not the Thai-highway study’s implementation.
Published method. A dedicated code repository and downloadable weights have not been verified.
Publication source ↗This study transforms discrete crash features into grayscale images using Weight of Evidence, then trains a VGG-based neural network to classify accident severity. It uses Thai highway crash records from 2011–2023.
Kasem Choocharukul & coauthors
Crash-record features encoded as images
VGG-based CNN + Bayesian optimization
Accident-severity classification
Evaluated with F1-score and AUC, with Bayesian hyperparameter optimization.
Published method. Public code and weights have not been verified. The authors state that they do not have permission to share the data.
Publication source ↗Researchers applied GeoSAM to extract building footprints from UAV true orthophotos with a 5 cm ground sampling distance. The study covers science and engineering areas at Chulalongkorn University.
Thepchai Srinoi; Thirawat Bannakulpiphat; Phisan Santitamnont
UAV true orthophotos · 5 cm GSD
Geospatial Segment Anything Model (GeoSAM)
Building footprints
Compared extracted footprints with manually delineated outlines using Intersection over Union (IoU).
The paper cites SamGeo (Wu & Osco, 2023). This is the upstream package, not the study’s code or data.
Published application of GeoSAM. A study-specific repository and downloadable weights have not been verified.
Publication source ↗A convolutional network predicts text, offset and border masks to locate curved and closely spaced text. Polygon offsetting and border augmentation help separate neighboring text instances in natural images.
Thanarat Chalidabhongse & coauthors
Natural-scene images
ResNet-50 + FPN
Text-region polygons
Evaluated on ICDAR 2015, 2017-MLT, 2019-MLT and Total-Text benchmarks.
Published method. Public code and downloadable weights have not been verified.
Publication source ↗Journal articles, conference manuscripts and preprints are identified by their source records. Years and authorship follow the original papers.
AI research directions
Four proposed directions building on published expertise. Project teams, datasets and timelines have not yet been confirmed.
Can a model flag cases that need an engineer’s review when cameras, lighting or road conditions change?
Road images and verified damage labels, separated by location and capture time; access and annotation rights must be confirmed.
Compare a detection baseline with confidence calibration and referral thresholds on held-out locations. Have engineers review flagged cases.
Precision and recall by condition, missed-damage rate, calibration error, and the fraction of images referred to people.
An evaluation protocol, error analysis and a reproducible model baseline.
Related researcher & published work ↗Does combining image-based condition estimates with engineering data improve maintenance choices under a fixed budget?
Road condition, traffic, repair costs and maintenance records, with documented coverage and missing values.
Compare current prioritization, AHP–TOPSIS and a data-driven alternative under identical constraints; analyse sensitivity to uncertain inputs.
Estimated condition improvement, maintenance coverage, budget compliance and stability of priorities; engineer assessment alongside model results.
A decision-support case study with traceable assumptions and comparisons.
Related researcher & published work ↗Can researchers find accurate answers faster when responses are grounded in approved documents?
Approved full texts, manuals and research notes; bibliographic metadata alone is not a full-text knowledge base.
Compare keyword search with document retrieval and cited answers on a researcher-written question set. Include questions with no answer in the corpus.
Answer correctness, citation support, retrieval recall, appropriate abstention and time to a verified answer.
A documented evaluation set and an internal prototype with page-level citations.
Method reference ↗Can suggested labels and targeted human review reduce annotation effort without lowering dataset quality?
Images or survey records with a small expert-checked reference set and a separate untouched test set.
Compare manual labelling with assisted labelling; select uncertain and unusual samples for review and track every correction.
Time per accepted label, correction rate, expert agreement and downstream model performance on the held-out test set.
An annotation workflow, quality report and dataset documentation.
Related researcher & published work ↗Links show relevant expertise or methods; they do not assign project leadership or claim completed results.
Field measurements, experiments and analytical models, from materials to road-management decisions.
2023 · Surveying & mapping
Compares onboard RTK, post-processed kinematic and precise-point-positioning workflows for UAV mapping using ground checkpoints.
2021 · Bridges & structural response
Combines a measured bridge influence line with iterative load estimation. Simulations and scaled experiments assess accuracy under road roughness and measurement noise.
2013 · Geotechnical engineering
Combines laboratory and field evidence to estimate small-strain stiffness parameters for Bangkok clay, supporting calibrated geotechnical finite-element models.
2018 · Water & environmental systems
Examines flow behaviour and treatment performance in a fluidised bed reactor, testing nitrate removal and anaerobic treatment of dilute wastewater.
2022 · Pavements & recycled materials
Repeated-load tests examined permanent deformation in recycled pavement materials treated with bitumen emulsion and cement. The laboratory results offer guidance for designing recycled pavement rehabilitation materials.
2017 · Transport & traffic operations
Field data from Bangkok’s Sathorn Road were used to model signal timing and reversible-lane operations in Synchro. The study compared proposed timings with observed police-operated signals to assess travel-time changes.
2025 · Road-maintenance decisions
Combines AHP factor weighting with TOPSIS ranking to prioritize road maintenance. Economic, engineering, environmental and social criteria help account for low-volume roads that traffic and roughness measures alone may overlook. The study evaluates the decision-support approach on a Thai rural-road network.
2022 · Rural-road priorities
Uses the Analytic Hierarchy Process to weight economic, engineering and social factors for rural-road maintenance. It relates priorities to the Department of Rural Roads’ strategic objectives, including network connectivity and the needs of low-traffic routes.
2020 · Local-road management
Assesses local-road management capacity across 442 local administrative organisations in Thailand. It examines road registration, condition and safety assessment, knowledge use, and maintenance equipment, identifying differences in capacity across types of local authority.
Each title links to the original publication.
Explore all 117 selected publications ↗Public repositories linked to the selected studies.
2026 · Pawarotorn Chaipetch
A pavement-distress detector informed by an audit of road-survey imagery. LearnableContrast adjusts local contrast, while EdgeSPD uses edge information during downsampling. The study compares five model scales with matched baselines; its findings are reported in an arXiv preprint.
2025 · Thanarat Chalidabhongse
Studies conformal prediction for quantifying uncertainty in HER2 status classification from breast-cancer immunohistochemistry images.
2024 · Pawarotorn Chaipetch
A lightweight pavement-damage detector based on YOLOv9 with partial attention. The study evaluates accuracy and inference speed using road images from multiple countries and an expanded set of damage classes. The journal article and its earlier arXiv version are grouped as one work.
2023 · Thirawat Bannakulpiphat; Phisan Santitamnont
Compares onboard RTK, post-processed kinematic and precise-point-positioning workflows for UAV mapping using ground checkpoints.
Find publications across transport, structures, water, geospatial engineering and computing.
2023 · Conference paper
Co-authored by Boonchai Sangpetngam · NCCE 28
The study used HDM-4 and data from Highway 344 to model pavement deterioration and maintenance costs over 20 years under four truck categories.
NCCE 28 · 9 July 2023 ↗2023 · Conference paper
Co-authored by Boonchai Sangpetngam · NCCE 28
The proceedings list research on flexible pavement strength, deflection, structural number and prediction models. The source provides a downloadable paper; its web abstract is not supplied.
NCCE 28 · 9 July 2023 ↗2025 · Research translation & road safety
Kasem Choocharukul · ThaiRAP / Faculty of Engineering, Chulalongkorn University
iRAP's January 2025 report identifies Kasem Choocharukul as ThaiRAP lead and describes road assessment and training with the Bangkok Metropolitan Administration. This is a programme activity record, not a journal publication.
iRAP · 22 January 2025 ↗Activity sources checked 8 September 2026. Dates and affiliations follow the original records; these activities are distinct from the proposed research direction.
