COMPARISON OF CLOSE-RANGE STRUCTURE-FROM-MOTION MODEL FROM DSLR AND LIGHTWEIGHT WEARABLE CAMERA FOR BRIDGE INSPECTION IN MYANMAR

摘要

Bridges are crucial in connecting communities and enhance economic activities. As of 2013, Myanmar has total of 4,728 bridge of which 741 bridge were classified as major bridge. Present geometric information is critical for inspecting bridge deformation and identifying suitable maintenance. Physical measurement methods are commonly used, yet intensive time and skilled human resources are required. Structure-from-Motion (SfM), using digital images to reconstruct 3D stage of structure, is possibly an economical alternative. Today various digital cameras are available. This study aims to compare SfM-based reconstructed model from DSLR and light-weight wearable action camera (both still image and video), as well as two software packages. Data was collected from Pathein Bridge, a suspension bridge with main span of 428m located in Pathein, Myanmar. In order to validate the accuracy of SfM models, point cloud from Terrestrial Laser Scanner (TLS) was used for relative quality comparison. Both SfM models were transformed into the same coordinates with TLS. First, cross-section profiles of all clouds are compared. Second, statistical median and interquartile range (IQR) of surface roughness is calculated and compared. Lastly, best-fitting primitive were generated for tower and pillar and their dimension are measured for comparison. The result implies that the quality of the models is highly dependent on image resolution and distance from camera sensor. In addition, different software affects the quality of the output. DSLR provided highest quality dense cloud, followed by GoPro in still image mode and GoPro in video mode. GoPro, which is a wearable camera, should be further studied because of its portability during field operation.