As the Building Information Modeling (BIM) is progressively integrated in the construction process, more and more actors are interacting with each others, the produced models gathering all information related to the building achievement. The BIM construction process relies on a collaborative work, each department (architecture, methods, structure) contributing to the different properties of objects found in this 3D representation of the future building. This working environment holds back vast amount of information, spread in highly complex data structures, with a high variability from models to models induced by the novelty of the process. All these issues and the absence of standard procedure are delaying the wide adoption of BIM in the industrial world. To tackle these issues, we propose 2 contributions: first, a method to automatically classify 3D objects extracted from the BIM models into different classes related to construction departments. Second, based on the classification of the BIM object, an adapted visualization of the whole model is proposed to the user depending of its department. By modifying the visualization properties, we bias the image saliency to render the relevant parts of the model more visually attractive than the irrelevant objects, without masking them. We propose an adaptation of the saliency estimation to fit the particularity of architectural images. This system is deployed in real working environment and its acceptation and usage is evaluated.
Keywords
Building Information Modeling, Virtual Reality, Unsupervised Learning, Architectural Saliency, Building Visualization