| dc.description.abstract | Using unmanned aerial vehicle (UAV) to assist bridge inspection has become a popular trend in Taiwan. However, since same type of bridge components are similar, it is very difficult to identify the correct position of a deterioration from photos or images captured by the UAVs. Thus, this study aims at generating three-dimensional (3D) models for bridges to facilitate the inspection task using those two-dimensional (2D) images.
Since there are many types of bridge components, the flying routes of the UAV are much more complicate than those for filming land slopes or buildings. In addition, photos with sky or clouds in the background of the under-bridge components such as slabs or piers, often make it very difficult to generate 3D bridge models. The same task is much easier for land slopes or buildings. After many tests in this study, it was concluded that a better 3D bridge model can be obtained under the following conditions: (1) capturing images under soft daylight such as in cloudy days, (2) keeping the camera have a 45 degree with the component, (3) avoiding capturing images of the UAV components, (4) capturing images by a single camera, (5) correcting fish-eye effect images, and (6) overlapping of images with 80% to 90% vertically and 50% to 60% horizontally.
Despite of numerous tests, the 3D bridge models generated in this study are not as precise as expected nor sufficient for inspection use due to the limited capabilities of current UAV, camera, and 3D generation software. However, the research results of this study still provide useful guidance for future study when capabilities of these hardware and software are significantly improved. | en_US |