Dissertation Defence: A Hybrid Framework for Traffic Sign Damage Detection and Condition Assessment Using Mobile Laser Scanning and Imagery

Ahmed Khataan, supervised by Dr. Suliman Gargoum, will defend their dissertation titled “A Hybrid Framework for Traffic Sign Damage Detection and Condition Assessment Using Mobile Laser Scanning and Imagery” in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Civil Engineering.

An abstract for Ahmed Khataan’s dissertation is included below.

Examinations are open to all members of the campus community as well as the general public. Registration is not required for in-person exams.

Abstract

Traffic signs are critical components of roadway infrastructure, yet traditional inspection practices remain largely manual, labour-intensive, and subjective. Although the past decade has seen a surge in using mobile mapping technology for automated data acquisition (i.e., inventory of traffic signs), existing approaches often fail to jointly assess structural integrity, functional visibility, and surface condition. This thesis addresses these limitations by proposing a hybrid framework for automated traffic sign damage detection and condition assessment that integrates mobile LiDAR and imagery for a comprehensive digital assessment of traffic signs.

The research develops three complementary methodologies. First, a LiDAR-based geometric change detection algorithm is introduced to identify structural damages such as deformation and misalignment. A sensitivity analysis evaluates the impact of point cloud density, validating the feasibility of using cost-effective LiDAR sensors. Second, a voxel-based visibility analysis framework is developed to quantify visual occlusions along a driver’s trajectory, automatically distinguishing between impediments caused by vegetation versus roadway geometry. Third, to address surface-level degradation, an image-based deep learning framework is proposed. To overcome the scarcity of real-world training data for damaged signs, the study implements a novel synthetic data generation strategy. This synthetic data is used to train Vision Transformer models enhanced with Low-Rank Adaptation (LoRA), enabling robust multiclass damage classification and severity assessment.

The thesis concludes by integrating these multi-modal outputs into a unified decision logic for maintenance prioritization. The results demonstrate that synthetic data significantly improves model generalization and that the proposed hybrid system provides a scalable, objective alternative to manual inspections. This research advances infrastructure asset management by delivering a holistic tool for ensuring the physical and functional reliability of traffic signage.
This research advances infrastructure asset management by delivering a holistic tool for ensuring the physical and functional reliability of traffic signage. Key terms associated with this research include traffic sign inspection, condition assessment, mobile LiDAR, visibility analysis, deep learning, synthetic data, Vision Transformer, and infrastructure asset management.