| Abstract |
Urban road waterlogging detection is a critical task for smart city management, requiring efficient and accurate solutions. Every year during the heavy rainfall season, urban road waterlogging is a recurring problem that severely impacts traffic operations, causing significant economic losses. This study proposes an edge computing framework that leverages the YOLOv8 model to address these challenges. The model was trained, optimized, and converted for deployment on the RDK X3 edge platform, enabling real-time processing of camera images for waterlogging detection and segmentation. To improve the performance of YOLOv8 in RDK X3, the feature decoding process was offloaded to post-processing, reducing computational overhead and enhancing inference efficiency. Field deployment tests, along with evaluations on a real-world dataset, demonstrated the effectiveness of the proposed framework, achieving high accuracy and robust performance in practical scenarios. This study highlights the potential of edge computing for enhancing urban resilience through intelligent waterlogging monitoring systems. Major findings include: (1) The model was trained and evaluated using a dataset of 3,956 annotated images and achieves waterlogging segmentation performance in real-world field tests, with 81% accuracy, a 74% F1 score, and 62% mIoU. (2) It offers low-latency waterlogging identification, critical for real-time urban management. (3) An innovative cloud-edge integrated method for water accumulation identification is proposed, enabling efficient urban waterlogging monitoring. © The Author(s), under exclusive licence to Springer Nature B.V. 2025. |
