Generating bridge geometric digital twins from point clouds

The automation of digital twinning for existing bridges from point clouds remains unsolved. Extensive manual effort is required to extract object point clusters from point clouds followed by fitting them with accurate 3D shapes. Previous research yielded methods that can automatically generate surface primitives combined with rule-based classification to create labelled cuboids and cylinders. While these methods work well in synthetic datasets or simplified cases, they encounter huge challenges when dealing with realworld point clouds. In addition, bridge geometries, defined with curved alignments and varying elevations, are much more complicated than idealized cases. None of the existing methods can handle these difficulties reliably. The proposed framework employs bridge engineering knowledge that mimics the intelligence of human modellers to detect and model reinforced concrete bridge objects in imperfect point clouds. It directly produces labelled 3D objects in Industry Foundation Classes format without generating low-level shape primitives. Experiments on ten bridge point clouds indicate the framework achieves an overall detection F1-score of 98.4%, an average modelling accuracy of 7.05 cm, and an average modelling time of merely 37.8 seconds. This is the first framework of its kind to achieve high and reliable performance of geometric digital twin generation of existing bridges.