Most building projects exist in three versions: the building that was designed, the building that was built, and the "as built" record of the building. Having the as-built version not quite reflect the building "as built" has long been a problem that has existed in part because collecting progress data during a project is a labor-intensive and costly effort. But thanks to the latest research with jobcams, some serious progress is under way toward solving this problem.
In recent years, owners and contractors have found camera-on-a-pole setups helpful. Connected to the Internet, jobcams are especially helpful in monitoring work without having to make a jobsite visit. Nonetheless, jobcam use as a tool for measuring real job progress has been virtually non-existent.
Now a research team at Stanford University is out to remedy this shortcoming. Thanks to the latest developments in computer vision and image processing techniques, the team is creating a low-cost system for automating the collecting of construction-progress data. Using time-lapse photos taken by jobcams, the system compares the images to a 3-D model of the project and its construction schedule (combined, this is also known as a 4-D model). Differences uncovered are automatically posted to the 4-D model, which becomes an accurate, up-to-date digital record of the project as built.
Converting pictures into data is nothing new. Optical character recognition (OCR) has been used for decades to convert a printed text page into digitized characters. And printed construction drawings have also been scanned and converted to vector data on a regular basis.
But applying computer vision techniques to jobcam images, project schedules, and 3-D CAD models is light years ahead of OCR and scanning. According to the project's principle investigator, Martin Fischer, time-lapse jobcam photo analysis will provide frequent, rapid, and automatic feedback on the performance of construction crews with respect to the project schedule.
Under the eye of the camera
Fisher's post-doctoral student Ragip Akbas, who is doing the lion's share of the work on the project, says the system will be applicable to multistory building construction and other projects where the construction process is in a bounded area that can be covered with jobsite cameras.
Configuring the number and locations of cameras that are sufficient given the site constraints is the first step in the process. Akbas adds that because his system would not be reliable from a single image and just a 3-D model, a sequence of time-lapse video images taken throughout the project will be collected. Then, on the computer, the segmentation step will follow.
A standard approach in computer vision, segmentation divides an image into regions based on the similarities of the pixels. Software that Akbas is developing will first analyze these regions, looking for changes. It will then automatically match the information with the 4-D model.
Automated data collection
The practice of automating data collection on construction projects is one pursuit that regularly demonstrates groundbreaking changes for the industry. The biennial World Conference on the Use of Automatic Data Collection in Construction has for over 10 years attracted a virtual who's who of technology implementers in construction. Organized by Eastern Carolina University's Connie Ciesielski, a construction management professor, the conference has repeatedly included representatives from such industry leaders as Bechtel, Fluor, Black & Veatch, CH2M Hill, and others. An overview of the most recent conference, held last year in Las Vegas, is online at www.sit.ecu.edu/cm-dept/adcic1_2003. The next conference is not yet scheduled.
Stanford's Fischer, an associate professor of engineering and director of the university's Center for Integrated Facility Engineering, has been working on the theories, concepts, and tools associated with 4-D models for about 10 years. Akbas's hope is that his system will also further extend the usefulness of existing 4-D models and jobcams, as well as encourage more designers in the industry to model projects in 3-D and 4-D.
In 2002, Fisher and Akbas collaborated on a prototype software development project called parametric 4-D modeling. The prototype generated a detailed 4-D model with specific information on where and when each crew on a project was scheduled to perform its work, based on a 4-D model generated from the project's master schedule. When there were crew conflicts, the system enabled users to automatically change a project's design to match the needs of the schedule.
But this prototype lacked an automated method to feed back actual progress data to update the 4-D model. Akbas says his progress-data collection system will be able to use the information from the detailed 4-D model to help determine activities in progress in the jobcam pictures, to determine quantities for the work in progress and completed, and to compare the progress in the jobcam pictures to the scheduled work.
While the project began last October, Akbas took some time off to work as a computer consultant specializing in 3-D models for Strategic Project Techniques, a San Francisco construction management consultancy. However, in May, with a September 2004 deadline looming to have an operational system delivered as the result of his research, he left the firm to devote his energies full time to its development. The research team is aggressively searching for an active project with a 3-D model, a schedule, and a jobcam that it can use to further test its system.
