Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6695395 | Automation in Construction | 2018 | 20 Pages |
Abstract
Various new data capturing technologies and object recognition systems have been developed to construct as-is building information models (BIMs) for operations and maintenance (O&M) management of existing buildings. However, a crucial challenge occurs in existing systems when semantic building information is captured under uncontrolled environmental conditions, especially in complex environments with poorly textured features (e.g., no obvious characteristics, edges, points, or lines). This study presents a semiautomatic image-driven system to recognize building objects and their materials and reviews the state-of-the-art object and material recognition methods and systems. A novel semiautomatic image-driven system was developed according to the new neuro-fuzzy framework for recognition of building objects and based on material classification procedures supported by an extensible texture library constructed to recognize their surface materials. More than 600 images were collected for the training process to develop this system, and more than 200 images were used for system verification. The results of the verification experiments show that the developed system can successfully recognize five kinds of building objects (i.e., beams, columns, windows, doors, and walls) and their corresponding surface materials from a single image taken by a handheld digital camera. Furthermore, the recognized building objects are automatically represented in industry foundation classes (IFC), a standard data schema for BIMs. The developed system is highly accurate, robust, and time-efficient for constructing as-is BIM objects in IFC and can help both BIM researchers and practitioners to develop information-rich BIMs in the O&M phase.
Related Topics
Physical Sciences and Engineering
Engineering
Civil and Structural Engineering
Authors
Qiuchen Lu, Sanghoon Lee, Long Chen,