Dam health monitoring is a critical component of dam management, requiring regular inspection, assessment, and maintenance to ensure structural integrity and operational safety. While 3D models of dams can be constructed given sufficient data, maintaining their up-to-date accuracy is challenging. Constructing an entirely new 3D model is typically resource-intensive due to the need for extensive data acquisition and computational effort. To improve efficiency, it is preferable to update only the regions of the model that exhibit actual changes, as unnecessary updates may introduce additional uncertainty. This paper proposes a novel change detection method, YOLO-DCD (YOLO-based Damage Change Detection), designed to support efficient 3D model updating using a single input image. A real image is first acquired, and its spatial location within the existing 3D model is estimated using a pixel-level localization technique. Based on the established pixel correspondences, the camera pose is computed using the Perspective-n-Point (PnP) algorithm combined with Random Sample Consensus (RANSAC) for robustness. A rendered image is then generated within the 3D model using a refined camera pose, adjusted to ensure complete coverage of the observed region in the real image. Damage detection is subsequently performed on both the real and rendered images using a YOLO(You Only Look Once)-based detection model. The comparison between the two detection results enables the identification of potential damage-related changes, which are categorized as unchanged regions, newly formed cracks, or new spalling. The proposed YOLO-DCD framework is evaluated using the Upriver Dam dataset, with experimental results demonstrating that the method achieves accurate and reliable detection of structural changes, facilitating efficient and targeted 3D model updates.
Zhao et al. (Tue,) studied this question.