SBIR Phase I: Automatic Generation of As-Built BIM Models Using a Single Video Camera

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to transform commercial construction through the democratization of scanning and automation of Building Information Modeling (BIM). Constant design and change management is a unique challenge that is costing the US construction industry billions of dollars annually. BIM allows significant reduction of these costs by facilitating an early analysis and decision-making process. The existing manual as-built BIM modeling technologies are only viable for large-scale commercial construction projects because the process is too laborious and expensive; in small projects, the projected savings hardly justify adopting the process. The proposed Scan-to-BIM technology has the potential to eliminate these barriers. Its ease-of-use, reduced file size, and automated object detection enable more scans and create more robust data models, unlocking the latent potential of the BIM process. Enormous value could be generated due to streamlining and data management capabilities that are offered by the technology in the form of accelerated project schedules, reduced number of requests for information and change orders, and timely identified design clashes. This paves the way to increasing the competitiveness of the US construction industry, by enabling new services that support automatic BIM modeling and effective infrastructure management. This Small Business Innovation Research (SBIR) Phase I project will investigate the technical and commercial feasibility of an automatic solution for the Scan-to-BIM problem in the construction industry. Scan-to-BIM refers to 3D scanning of a physical structure (in the form of a dense point cloud) and converting the raw spatial data into an object-oriented, semantically-rich model. Although there are a few technologies that aim to facilitate this process, the 3D modeling component remains primarily a manual and time-consuming task. When using the proposed technology, if proved to be feasible, a non-technical worker videotapes the target structure. The video is automatically sent to a server for processing. A unique videogrammetric framework reconstructs the 3D geometry of the structure. This generates a dense point cloud, an edge cloud, and a line cloud (which are still raw but more meaningful spatial data). It then extracts object hypotheses from key frames and spatial data in the form of observations. Afterwards, it optimizes a joint probability distribution function and infers a 3D layout. The outcome is an object-oriented 3D representation of the structure with embedded geometrical attributes and property sets. This advances the state-of-the-art in robust intelligence, particularly the interpretation of complex, unstructured data.