Institutional-Repository, University of Moratuwa
Welcome to the University of Moratuwa Digital Repository, which houses postgraduate theses and dissertations, research articles presented at conferences by faculties and departments, university-published journal articles and research publications authored by academic staff. This online repository stores, preserves and distributes the University's scholarly work. This service allows University members to share their research with a larger audience.
 Research Publications |  Thesis & Dissertation |  E- Books |
 UoM Journal Publications |  UoM Conference Proceedings |  Articles published in Scimago's Q1 journals |  UoM Research Reports |  Other Articles authored by UoM staff |
Recent Submissions
item: Thesis-Full-text
Effective management of learner integrity in online assessments : an approach using keystroke dynamics
(2026) Seneviratne, HAAKM; Perera, I; Fernando, S
Educational institutions worldwide face a serious issue when conducting online assessments due to certain candidates cheating. Review of related work in the domains of learner integrity and technology assisted fraud prevention indicated gaps in the application of technology for maintaining learner integrity during online assessments. This thesis addressed these gaps by conceptualizing a system for the effective management of learner integrity during online assessments. The use of the behavioral biometric called ‘keystroke dynamics’ was thoroughly reviewed and how it could be adapted to authenticate candidates continuously and non-obtrusively while they take online assessments with no accessories deployed at the remote site was conceptualized. The thesis describes how the proof of this concept was established through experiments. This includes the setting up of a fictitious online testing institution and deploying a group of volunteers to roleplay as candidates. Tests were conducted online to simulate a structured initial registration to acquire biometrics for profile creation. Novel features extracted from the acquired biometrics were used to train anomaly detection models employing a variety of binary classifiers. This was followed by online tests that simulated genuine test taking as well as fraudulent test taking. These scenarios were played back offline while continuous authentication was exercised using the trained classifiers. Both conventional metrics and metrics applicable to continuous authentication that have been established recently were used for the performance assessment of the classifiers. Quantitative analysis of the test results indicate that the system is capable of detecting impersonation with high accuracy while other fraudulent acts are also detected with dependable accuracy. The system is also capable of the absolute and profile independent detection of clipboard transfer of text during online assessments. The thesis also discusses schemes for improving the overall detection capability of the system and suggests several directions to inspire future work.
item: Thesis-Abstract
Use of deep learning to investigate the relationship between tortuosity of the vasculature of the retina and the bulbar conjunctiva
(2026) Naim, AM; Liyanaarachchi, R; De Silva, A; Dayananda, N
Accurate segmentation of conjunctival vessels is critical for the non-invasive assessment of systemic microvascular disease. While diabetes is commonly evaluated using retinal imaging, alternative accessible biomarkers remain underexplored. Conjuncti val vessels are directly visible and may reflect systemic vascular changes; however, reliable quantification is challenging, as conventional deep learning models prioritize pixel-wise accuracy without preserving vascular topology or diameter-specific structure. In this thesis, we propose a transformer-augmented U-Net architecture designed for topology-preserving, diameter-sensitive segmentation of conjunctival vasculature from standard DSLR-captured images. The model integrates dilated convolutions, attention mechanisms, and transformer-based contextual modeling to balance local feature representation with global structural awareness, enabling improved detection of small-caliber vessels while maintaining vascular continuity. A curated dataset of conjunctival images with diameter-specific annotations was developed for training and evaluation. Through systematic ablation studies, we demonstrate that no single architectural component is sufficient in isolation; rather, the full hybrid design provides the most consistent preservation of vessel geometry. Compared with multiple state-of-the-art segmentation methods (U-Net, Attention U-Net, Tran sUNet, IterNet, and U2-Net, Swin-UNet), the proposed approach achieves comparable pixel-wise performance while significantly reducing tortuosity error and centerline deviation, highlighting the importance of topology-aware segmentation for downstream analysis. Clinically, the proposed framework enables robust estimation of vessel tortuosity, revealing significantly elevated conjunctival tortuosity in individuals with diabetes compared to healthy controls, consistent with established retinal findings. These re-suits support conjunctival imaging as a low-cost, non-invasive modality for microvas cular phenotyping and suggest its potential for scalable screening and longitudinal monitoring. Despite limitations related to dataset size, patch-based training, and cross-domain generalization, this work establishes that topology-preserving, diameter-aware segmentation is essential for clinically meaningful vascular analysis. Future work will focus on expanding multi-institutional datasets, improving spatial calibration, and validating the framework across diverse populations and vascular pathologies.
item: Thesis-Abstract
Experimental and numerical investigation of reinforced concrete dapped-end connections with neural network-based optimisation
(2026) Dharmawansha, DAST; Rajapakse, RMCM; Herath, HMST; Fernando, PLN
This study focuses on both an experimental investigation and a data-driven approach to advance the understanding and design of dapped-end connections, which are widely used yet structurally vulnerable elements in precast concrete buildings and bridge structures. The first part of the study presents an experimental investigation conducted on four large-scale dapped-end connection specimens to examine the influence of bond deterioration on strength and service behaviour. Bond deterioration near the re-entrant corner was simulated by de-bonding a 100 mm length of either the hanger or flexural reinforcement using a plastic sleeve. Full-field Digital Image Correlation measurements were employed to capture deformation and crack kinematics, which were used to quantify key resistance mechanisms. Detailed deformation measurements enabled an in-depth discussion of specimen behaviour. Each specimen was modelled using the VecTor2 finite element software to further examine its response. The finite element models predicted peak resistance with an average experimental-to-predicted ratio of 1.01 and a coefficient of variation of 7.34%. The second part of the study focused on data-driven approaches. Although previous studies have investigated dapped-end connections using analytical and experimental methods, the application of machine learning remains limited due to the scarcity of experimental data, which constrains predictive accuracy and model generalisation. To address this limitation, a novel framework combining numerical simulation with Conditional Tabular Generative Adversarial Network (CTGAN) -based data augmentation was proposed. A numerical database of 720 cases was developed and expanded to 1,400 instances using CTGAN. Artificial Neural Network (ANN), Genetic Algorithm-ANN (GA- ANN), and Particle Swarm Optimisation-ANN models were trained and evaluated. Validation using 64 unseen experimental datasets demonstrated improved model generalisation. shapley additive explanations analysis indicated that GA-ANN predictions were consistent with deformation compatibility principles. Finally, a machine learning-assisted design model was developed to provide multiple solutions for a given design problem, supporting the optimisation of dapped-end connection design.
item: Thesis-Full-text
Development of a friction prediction model of asphalt concrete at the construction stage based on Sri Lankan aggregates
(2026) Thenuwara, HV; Pasindu, HR
The skid resistance at the tire-pavement interface ensures adequate braking force for road user safety. Insufficient friction at the wearing course emerges as the primary pavement-related factor of highway crashes. Governing factors of pavement friction include pavement properties, vehicle/tire/driver conditions, and environmental conditions, with pavement characteristics, particularly the mineralogical and textural attributes of surfacing coarse aggregates exerting dominant control. Coarse aggregate polishing resistance critically determines the available skid resistance of an asphalt concrete road, prompting extensive research on aggregate’s friction retention capacity under traffic loading. Hence, globally a number of pivotal studies have been conducted to explore polishing behaviour of roadstones and to develop prediction models to forecast skid resistance capacity of flexible pavements. In Sri Lanka, road safety garners increased emphasis amid expressway network integration, yet studies on local aggregate friction properties remain limited, with fewer user-safety guidelines incorporated in the national Standard Specifications for Construction and Maintenance of Roads and Bridges (SSCM-RB). This study thus evaluates friction characteristics of indigenous Sri Lankan road surfacing aggregates via the Polished Stone Value (PSV) test (employing Accelerated Polishing and British Pendulum) alongside petrographic and microscopic analysis of polishing behavior in order to develop a friction prediction model for national asphalt concrete pavements.
item: Thesis-Full-text
Vision-based accident anticipation with deep neural networks
(2026) Vipooshan, V; Chitraranjan, C
Traffic accidents remain a leading cause of fatalities and injuries worldwide, making the ability to anticipate hazardous situations in advance critical for improving road safety. Accident anticipation systems enable timely preventive actions, such as driver alerts and collision avoidance maneuvers, and play a key role in advanced driver assistance systems (ADAS) and autonomous driving. While many existing approaches rely on multiple sensors such as LiDAR, radar, and GPS, using dash-cam videos offers a cost-effective and easily deployable alternative, albeit with significant challenges due to the complexity of real-world driving environments and real-time constraints. In this thesis, we investigate efficient and accurate accident anticipation from dashcam videos through two complementary approaches. First, we propose an enhanced spatio-temporal graph-based framework (STAGNet), which improves accident prediction by incorporating richer spatio-temporal features and aggregating them using recurrent neural networks. Second, we introduce VAGNet, a novel architecture that leverages global scene representations instead of computationally expensive object-level features, combining transformer and graph modules with the vision foundation model VideoMAE-V2 for efficient feature extraction. Extensive experiments conducted on multiple benchmark datasets, including DAD, DoTA, DADA, and Nexar, demonstrate that the proposed methods achieve superior performance in terms of average precision and mean time-to-accident compared to existing approaches. Notably, the global feature-based framework significantly improves computational efficiency while maintaining high predictive accuracy. Together, these contributions advance the development of scalable, real-time accident anticipation systems using only dash-cam video, providing a practical pathway toward safer intelligent transportation systems.