As-Built Modeling of the Physical World Using Video

This research thread seeks to build faithful models of physical objects observed in the real world. By focusing on physical objects, the research seeks to differentiate from standard Structure-from-Motion or SLAM techniques through the exploiting the compact nature of the object's support. Any additional geometry or information that is available is also used.

Application: Civil Engineering

Application of SfM/SLAM to civil engineering is motivated by the grand challenges of engineering as reported by the National Academy of Engineering (2008) in the document "Restoring and Improving Urban Infrastructure." Modeling and annotating the as-built nature of existing infrastructure is a time consuming task. While it may seem beneficial to avoid this task, doing so leads to incomplete quality assessment during construction and increases maintenance costs downstream. Establishing an automated system for generating as-built building information models (BIM) from visual measurements of as-built infrastructure would eliminate the current bottleneck and enable more complete and efficient operations. Research in this area has included the investigation of SfM/SLAM techniques, their accuracy, and their modification to suite civil engineering needs and practice. Current research seeks to move from point cloud data to BIM automatically via machine learning for cost-effective generation of as-built models.

Students: Abbas Rashidi (co-advised), Guangcong Zhang.

Support:

This research was supported in part by the National Science Foundation (#1031329). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.