Abstract

Vietnam is a country where motorbikes are the primary mode of transportation, with millions of people commuting daily on two wheels. However, the safety of these road users is significantly threatened by damaged or missing drain covers. During rainy seasons, wet and slippery road conditions combined with height differences between the drain cover and the road surface often lead to serious accidents. Pedestrians, cyclists, and motorcyclists near the roadside are especially vulnerable to such hazards. In recent years, Ho Chi Minh City and many other regions in Vietnam have recorded numerous injuries and fatalities resulting from these damaged infrastructures. Conventional inspection methods for detecting drain cover failures are mainly manual, requiring considerable time, manpower, and cost. These methods are not efficient enough to ensure timely detection and repair. Therefore, developing an automated and intelligent system for early identification of damaged drain covers is essential to improving urban road safety. This study proposes an enhanced detection approach based on computer vision and unmanned aerial vehicle technology. Specifically, the YOLOv8 deep learning model is employed to identify and classify damaged drain covers from aerial images. A new dataset is constructed and annotated to train and evaluate the model under various real-world conditions. The proposed system enables efficient large-scale monitoring, minimizes human effort, and supports authorities in taking prompt maintenance actions. The results of this research are expected to contribute to safer transportation environments, reduce accident risks, and demonstrate the potential of integrating artificial intelligence and unmanned aerial vehicles in urban infrastructure management. In the future, this project will be further developed by expanding the dataset, improving detection accuracy in diverse lighting and weather conditions, and integrating real-time processing to support continuous monitoring and smart city applications.