Professor of Computer Science Michael Black has the decency not to undress you with his eyes. But he's more than happy to do it with his computer.
Black and Alexandru Balan GS have developed a program that allows a computer to determine the shape of a person's body based on images of the person wearing clothes.
The two developed the program to meet a need for realistic models of the human body. "We've been working on tracking humans in video for a number of years," Black said. "But until now, our models of the human body looked like the tin man from 'The Wizard of Oz.' "
The technology could have applications in several fields, including forensics and medicine.
Black started by looking at a model already used by the graphics industry. The problem with the existing model, he said, is that it can only determine the shapes of people wearing tight-fitting clothing.
But regardless of what one wears, everybody's shape is relatively constant, he said. So by combining images of people in different poses, the computer can "see" a person's body.
For a computer, an image is just a bunch of numbers, Black said. The computer's function is to derive information from that string of numbers.
"We had to teach the computer what people look like," he said. To do so, Black and Balan used detailed laser-range scans of 2,400 different people. Each person stood still in a scanner while lasers mapped out their bodies. Using the data gathered from the images, the pair then used statistical analysis to find the variations among people.
The result of their work is a program that allows the computer to "see" a person based on the input images and extract information about that person, such as height and waist measurements. The computer can determine a person's height to within a few millimeters of the actual value. Waist and chest sizes are more difficult to determine because they vary more based on the clothing somebody wears, so the program can determine someone's waist measurement to within one to two inches.
The research earned enough attention that Black and Balan were invited to present it at last month's 10th European Conference on Computer Vision in Marseilles, France.
Black said the University has filed for two patents on the technology, which has a multitude of potential uses.
Director of Intellectual Property Luann Cserr, who is overseeing the patent application process, said the technology has potential applications in forensics, fashion, graphics, gaming, film and even medicine. "It could enable doctors to be more accurate" in determining whether patients' body types puts them at risk for a disease, she said.
"People are getting really excited about using the program," Black said, adding that the Rhode Island State Police have been in touch with him about applying the technology to surveillance tapes to help catch criminal suspects.
"Obviously, suspects robbing a convenience store aren't going to do it naked," he said. The new technology will allow the police to determine the proportions of the person responsible for a crime, even if the suspect is wearing a mask or baggy clothing.
Currently, in order to determine a person's height and weight from video footage, the police must measure the scene of the crime and determine the person's proportions based on the size ratio of the person to the surrounding environment, said Lieutenant Dennis Pincince of the Rhode Island State Police. He said the new technology will make identifying suspects much easier because it will allow the police to take the suspect's measurements directly from the video.
The technology could also be used in online shopping, Black said.
"The fashion industry is very interested in virtual try-on," he said. Such a technology would allow people shopping online to see what clothes look like on their bodies without having to try them on in person.
As for negative consequences of such a technology, Black does not believe it is harmful. "I don't see that there are major privacy concerns because it's not actually an invasive technology," he said, adding that currently, all scans are voluntary.
In the meantime, there are still a few kinks to work out of the system.
"We had a subject who was wearing a baby-doll sort of top," Black said. "The computer thought she was much fatter than she actually was."
