Highly Automatic Approach to Architectural Floorplan Image Understanding & Model Generation

Authoring of a 3D model using computers typically involves tedious manual efforts. In this paper, we propose a highly automated approach to generate the 3D model of a building from its 2D floorplan image. The proposed system treats the floorplan as a vector image in that the lines in the plan are converted into vector outlines. Building wall identification is thus reduced to a matching problem of the outlines in the plan. A number of advantages are resulted from using this approach. For example, special constructs such as doors can be recognized through the geometric characteristics of the line signatures. The concept of interior space can also be explored through examining the line loop within the plan. Our system allows the generation of building interior as well as the exterior model. The presented algorithm is useful for data generation in architectural walkthrough of city scenes as well as 3D games development.