From Measured Geometry to Editable BIM: A Representation-Centered Review of Existing-Building Scan-to-BIM Reconstruction

Existing-building Scan-to-Building Information Modeling (Scan-to-BIM) transforms measured observations into editable Building Information Modeling (BIM) content with reliable geometry, coherent relations, semantics, and model structure. This remains difficult because point clouds, depth data, and photogrammetric geometry contain noise, occlusion, incomplete visibility, irregular sampling, and registration uncertainty, whereas BIM reconstruction requires stable objects, explicit relationships, semantic organization, and model context. Existing reviews have emphasized sensing, workflows, reconstruction tasks, and applications. Changes in representation and the continuity of building information have received less attention in cross-study synthesis. This review therefore adopts a representation-centered narrative synthesis organized by the building information made computationally explicit and the settings through which it is evaluated. Raster/projection, graph/topology, point-set/primitive, and vector-native/parametric/BIM-operation representations expose complementary aspects of observation regularization, spatial relations, metric geometry, object parameters, authoring operations, and model state. Evaluation settings expose different subsets of these properties, so reported performance remains tied to the output and reference structures used. Across the literature, representation evolution expands the building information made computationally explicit, but measurement support, object identity, spatial relations, semantics, uncertainty, and model state may change across representation interfaces. The validity and traceability of this information remain central to evaluating the transition from measured geometry to editable and maintainable BIM content.