Semantic Structure from Motion for Railroad Bridges Using Deep Learning

Current maintenance practices consume significant time, cost, and manpower. Thus, a new technique for is required. Construction information technologies, including building modeling (BIM), have recently been applied to the field carry out systematic productive planning, design, construction, maintenance. Although BIM increasingly being structures, its application existing structures has limited. To apply an structure, three-dimensional (3D) model of structure that accurately represents as-is status should be constructed each structural component specified manually. This study proposes method constructs 3D specifies automatically using photographic data with camera installed on unmanned aerial vehicle. procedure referred as semantic from motion because it point cloud together information. A validation test was carried railroad bridge validate performance proposed system. The average precision, intersection over union, BF scores were 80.87%, 66.66%, 56.33%, respectively. could improve current scan-to-BIM by generating specifying automatically.