DESIGN OF A DYNAMIC MONITORING AND TIERED EARLY WARNING SYSTEM FOR CULVERT WATERLOGGING BASED ON A CLOUD-EDGE COLLABORATIVE ARCHITECTURE

Abstract

In response to the frequent waterlogging of urban culverts induced by extreme rainfall, the vulnerability of traditional monitoring methods, and the lack of targeted early warnings, this paper designs and implements an Internet of Things (IoT)-based dynamic monitoring and tiered early warning system utilizing a cloud-edge collaborative architecture. At the perception layer, the system employs a non-contact computer vision approach. By deploying dual lightweight YOLOv8 object detection models on edge computing nodes (Raspberry Pi 4B), it synchronously extracts on-site water depth and the chassis height of passing vehicles in real time. At the decision layer, a tiered early warning logic based on dynamic "water level-vehicle type" matching is proposed. This approach overcomes the limitations of traditional uniform static thresholding strategies and enables sub-second local audio-visual traffic control directly at the edge. Furthermore, the edge nodes transmit lightweight structured data to the Huawei Cloud IoT platform for aggregation, ultimately delivering public services via a WeChat Mini Program. Users can leverage the embedded map component to monitor the real-time waterlogging status across various monitoring sites and receive vehicle-specific routing recommendations. Prototype testing results demonstrate that the system's visual perception error is maintained within a reasonable margin, and the overall response latency at the edge is less than 500 ms. Crucially, the system retains offline resilience and local warning capabilities even in extreme scenarios involving network interruptions. This research provides a low-cost, robust engineering solution for smart urban flood control and emergency traffic dispatching.