New Method for 3D Scanning Buildings Uses Machine Learning

New system automates the process of acquiring detailed building information from 3D scans.

As 3D scanning technologies continue to become more accessible, industries are finding new applications for using them. Among others, these include construction companies that are now using 3D scanners to acquire detailed building information. While many newer buildings may still have accurate construction data stored electronically for remodeling or refurbishing existing buildings, this isn’t always the case—particularly for older buildings built long before the advent of computer-aided drafting (CAD) and building information modeling (BIM).

In an effort to streamline the time-consuming and tedious task of collecting architectural details by hand, a team of Stanford researchers has recently added a level of automation to the existing process of collecting reliable building information with the aid of 3D scanners.

 

The system replaces the manual method of using a measuring tape and clipboard. (Photo: Stanford University School of Engineering.)

Spearheaded by Greek doctoral student Iro Armeni, the system utilizes existing 3D sensing technologies and light sensors to accurately measure features within a building interior through scanning each room into a large data file commonly known as a point cloud.

While a point cloud file can hold a mass of data about the points it has captured, it still needs to be converted into usable CAD data. This is typically an additional and fairly manual process. In Armeni’s system, the resulting data file is then fed into a computer where an algorithm automatically identifies both structural elements and interior furnishings, eliminating the need for a person to identify these items manually in the process of converting a point cloud file to CAD.

Armeni, who had previously been an architect on the Greek island of Corfu, came up with the concept after growing frustrated with her own experience of redrawing building plans with a tape measure that oftentimes ended up being inaccurate. 

In her early experiments, Armeni replaced her tape measure with laser scanners and 3D cameras to take measurements using light with high accuracy. The amount of time it took for a beam of light to hit another point within a room provided the basis for collecting millions of measurements that made up the resulting “raw point cloud” containing a building’s unique information.

To automate the process of identifying unique attributes of the building’s information, Armeni and a team of fellow Stanford researchers developed a computer vision system that makes use of machine learning to distinguish common architectural elements such as windows, doorways, halls and stairs.

“People have been trying to do this on a much smaller scale, just a handful of rooms,” said Silvio Savarese, a Stanford assistant professor in computer science. “This is the first time it’s possible to do it at the scale of whole buildings, with hundreds of rooms.”

The next step for Armeni’s team is to create a similar algorithm to help automate and track a building’s entire lifecycle from design to construction to eventual demolition.

As engineers, we shouldn’t lose time trying to find the current status of our building,” Armeni explained. “We should invest this time in doing something creative and making our buildings better.”