Speaker
Description
The joint applications of images and point clouds are underdeveloped in the realm of infrastructure operation and maintenance (O&M), limiting their potential for mutual reinforcement towards each other. In this study, a novel approach of image and point cloud data fusion for crack detection and location registration is proposed. High-quality images and point clouds are efficiently obtained simultaneously by a terrestrial laser scanner. Deep learning-based methods are utilized to automatically detect cracks from images, capturing crucial information regarding their shape, scale, and location. Subsequently, the detected crack information is updated onto the bridge point cloud through 2D-to-3D projection, enhancing its detail and accuracy. This method yields reliable as-is point cloud models, which serve as a critical foundation for advanced O&M tasks such as finite element analysis, building information modeling, and digital twinning.
