Proactive risk assessment and infrastructure safety management

◎Research at a glance

Over the past two years, I have developed expertise in training and applying AI-based models, particularly Computer Vision (CV) ones, alongside geospatial tools within a road safety pipeline. Certain attributes, such as pavement defects, benefit from well-developed automated extraction methods, yet remain underutilized in predictive transportation research. Infrastructure and environmental attributes have rarely been linked to Surrogate Measures of Safety. Our findings indicate that those attributes contribute modestly to harsh-event prediction relative to traffic and behavioral metrics, though they offer valuable supplementary insight. Additionally, automated labeling does not outperform manual labeling in predictive power, but it can reveal overlooked interactions among attributes.

In the following months, we aim to leverage CV models to extract attributes from open-source image databases and construct a complexity index. This index will be modeled against HE at intersections in central areas of Greece, with the goal of identifying which attributes most significantly affect driving smoothness.