Monitoring changes in river water levels is a critical component for enhancing the accuracy of flood prediction and mitigating disaster risks. However, conventional fixed water-level gauging systems involve high installation and maintenance costs and are often impractical in upstream areas or observation blind spots due to geographical inaccessibility. This study proposes an image-based water level estimation system that accurately calculates water levels from a single image captured by a smartphone. The proposed method integrates a pre-constructed 3D model, visual localization techniques, image-based water surface segmentation, and a ray casting algorithm to estimate the 6-DoF camera pose and compute the actual water height based on the detected water boundary within the image. To further enhance the accuracy of pose estimation, an additional correction step using iMU data and homography refinement is applied. Field experiments conducted at the Sebyeonggyo section of the Oncheoncheon Stream in Busan, South Korea, demonstrated a mean water level estimation error within 2 cm, validating the system's high accuracy. These results confirm the feasibility and reliability of the proposed approach. Leveraging commonly available smartphones, this participatory monitoring system offers a cost-effective and scalable solution to complement existing hydrological networks. With a flexible framework applicable to diverse river environments, the system demonstrates the practical viability of image-based water level estimation and provides a promising technological approach for flood management strategies in response to climate change.

E-mail: taeyun@pusan.ac.kr