Our work spans imaging metrology, sensor calibration, and 3D spatial data analysis.
Imaging metrology and sensor calibration
Rigorous geometric modelling, self-calibration, and systematic-error characterization of terrestrial laser scanners, digital and range cameras, and X-ray imaging systems. This work also develops the estimation theory and optimal measurement-network design that underpin precise 3D reconstruction.
Point cloud processing and geometric feature extraction
Registration, segmentation, and robust fitting of geometric primitives such as planes, cylinders, and ellipses from laser-scanning point clouds, including object and target recognition under noise and occlusion.
As-built and industrial reality capture
Automated 3D documentation, dimensional verification, progress monitoring, and condition assessment of industrial and construction environments, with applications to mechanical piping and structural elements.
Building and indoor 3D reconstruction
Automated extraction of architectural and interior structure, including storey separation, openings, and indoor objects, from complete laser scans for building modelling.
Learning-based airborne LiDAR analysis
Statistical and deep-learning methods, including maximum-entropy approaches, for classification, outlier removal, and quality enhancement of multispectral airborne LiDAR point clouds.
Biomedical and biomechanical imaging
Calibration and modelling of multimodal imaging systems, such as dual-fluoroscopic X-ray and range or RGB-D cameras, for human motion capture and biological measurement.
Cultural heritage and geological surface documentation
3D recording and digital preservation of at-risk cultural-heritage sites, and the extraction and analysis of geological surfaces such as rock-mass discontinuities from point-cloud and photogrammetric data.
Click a project to expand its full description, methods, and lead researcher.
UC3D is a smartphone-based 3D mapping platform that combines mobile sensors, LiDAR, and AI methods to generate accurate 3D models. The system is being applied to document and preserve heritage assets in collaboration with Parks Canada, supporting digital archiving, condition assessment, and disaster preparedness.
The project aims to provide a practical, low-cost alternative to traditional reality capture systems.
This project develops methods for detecting structural changes and surface deterioration from multi-temporal point clouds. By combining geometric analysis with AI-based feature extraction, the research supports heritage conservation, infrastructure monitoring, and damage assessment.
Multi-temporal point cloudsGeometric analysisAI feature extraction
Abdelrahman Abdelghany · PhD Student
Photogrammetric self-calibration is compared against checkerboard-based closed-form calibration across downstream tasks, including structure-from-motion reconstruction, monocular visual odometry, and novel view synthesis.
The evaluation is further extended to benchmark datasets, where photogrammetric self-calibration is compared against manufacturer-provided calibration parameters.
This thesis project develops an automated, non-invasive 3D reconstruction pipeline to measure reindeer antlers, addressing the limitations of previous measurement methods. By combining a multi-camera optical setup with Gaussian Splatting, the system overcomes data-quality challenges caused by animal movement and inconsistent lighting.
Initial laboratory testing yielded a mean cloud-to-cloud distance of 3.4 to 3.7 mm compared to reference laser scans, and the system is now deployed in the field for multi-season collection. Ultimately, this scalable pipeline will provide high-quality 3D datasets to advance interdisciplinary research across biodiversity, environmental monitoring, and regenerative medicine.
This project develops dual fluoroscopy systems for musculoskeletal biomechanics and osteoarthritis research, focusing on system calibration, 3D motion tracking, and the application of imaging technologies to capture human joint motion.
MedFlow investigates whether a plausible whole-body FDG-PET image can be synthesised from a routine CT scan alone, and where that synthesis is limited by CT itself rather than by the model. FDG-PET reveals metabolic activity decisive for cancer detection, staging, and treatment monitoring, yet remains scarce relative to the near-ubiquitous CT.
A 2.5D rectified flow model trained on the whole-body AutoPET cohort reproduces normal anatomy with high fidelity but systematically under-predicts per-lesion tumour uptake (SUV), which proves largely unrecoverable from CT appearance. Formalised through the data processing inequality, this establishes an information-theoretic ceiling on CT-only quantitative PET synthesis, reframing the shortfall not as a model failure but as a measurable limit of the modality.
Rectified-flow generative models2.5D synthesisAutoPET cohortInformation theory
Jackson Cooper · MSc Student
This project develops network design methodologies for mobile scanning systems used in indoor mapping applications. It focuses on optimizing data acquisition workflows, viewpoint planning, and spatial coverage and uncertainty to improve the quality and efficiency of indoor geospatial data collection.
The goal is to support more reliable and accurate indoor mapping solutions for a variety of built-environment applications.
Network designViewpoint planningIndoor mapping
Abdoul Karim Ouangre · PhD Student
This project investigates the integration of Real-Time Location Systems (RTLS) with mobile LiDAR platforms to improve positioning accuracy and overall point cloud quality. By combining complementary sensor technologies, the research aims to reduce drift and enhance the geometric reliability of mobile mapping datasets.
The work supports the development of more robust indoor and complex environment mapping workflows.
RTLSMobile LiDARSensor fusion
Abdoul Karim Ouangre · PhD Student
This research direction integrates LiDAR point clouds with Gaussian Splatting workflows to improve the geometric accuracy of 3D scene reconstruction. The objective is to leverage LiDAR-derived constraints during Gaussian generation, creating more accurate and scalable digital representations that can be captured using consumer devices such as smartphones.
Future applications include BIM and HBIM workflows, semantic scene understanding, and automated extraction of architectural elements through computer vision and segmentation techniques.
