The newest computer vision technology can't literally see through your sweater and slacks, but it can determine the shape of your unclad body.
Brown University computer scientists have developed software that takes ordinary images of people dressed in street clothes and digitally peels away the layers to determine their 3-D body shape.
The potential applications are many: Shoppers could create a customized, anatomically accurate avatar to see how well clothes would fit them without stepping into a dressing room; doctors could use a scan to understand how body shape plays a role in disease risk; people playing video games could upload their virtual dimensions as a character in the game.
The researchers are patenting their technology, and see potential uses everywhere from Hollywood special effects departments to sports medicine.
The software originated in efforts to solve a more low-tech challenge: fighting crime at the corner store.
In a partnership with Rhode Island State Police, Brown computer scientist Michael Black developed tools to figure out a person's basic physical attributes from a grainy surveillance video.
"They want to know how tall is this person, how much do they weigh? So we developed some simple tools," Black said. "But if we wanted to (get) a few other measurements . . . we have to deal with clothing." Or, more specifically, the figure under the clothing.
Using more than 2,000 laser scans of minimally clad people, Black and graduate student Alexandru Balan gave the computer information about what the human body looks like. Then, they used images of different poses - along with a model that factors in how clothing becomes looser and tighter when a person moves - to create nude digital mannequins.
As a test, they had six subjects pose in a variety of poses and outfits, and then used the computer program to predict measurements and gender.
They were able to predict gender accurately 94 percent of the time, and their calculations closely matched people's actual body shape. But it's not fool-proof - a trench coat tripped them up.
Stan Sclaroff, a computer science professor at Boston University, commended the technology: "This is the first work I've seen that can . . . come up with a reasonable estimate of (a person's) dimensions, without making them strip down."
Carolyn Y. Johnson can be reached at cjohnson@globe.com ![]()


