MIT

Researchers are developing algorithms that can generate "invisibility cloaks" for ugly objects like utility boxes.

Disguising eyesores like sidewalk utility boxes isn't usually a big priority for city governments. The work—if it's done at all—has sometimes fallen to enterprising local artists. Recently, however, computer scientists are working on a more high-tech approach.

Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory and a few collaborating institutions have developed an algorithm that can automatically generate camouflage coverings to hide physical objects. The algorithm essentially takes in a set of photos capturing multiple perspectives of a scene, analyzes them, and outputs a surface pattern that can camouflage an object placed in that setting. 

And the technique seems to be working. In a new study, the researchers timed how long it took human volunteers in Amazon’s Mechanical Turk system to detect a camouflaged box placed in different settings. The result? More than three seconds on average. That means you probably won’t notice the box in a passing glance.

Take a look at these examples—in each scene, try to find a virtual 3D box that’s been covered  with patterns generated from the algorithm. The video below shows even more examples. 

Courtesy of Andrew Owens/CSAIL
Courtesy of Andrew Owens/CSAIL

According to Andrew Owens, an MIT graduate student and the project’s lead researcher, it’s possible to make an object essentially invisible from a single viewpoint. The challenge his team is trying to tackle, however, is generating camouflage patterns that work wellor well enoughfrom all angles. In trying to do so, the current algorithm has to decide which detection cues (such as how well the object’s outline blends into the background, or how distorted the object looks) are more important to conceal in a certain scene.

Additional challenges moving forward are revealed in the team’s preliminary test with a real-life box, placed as pictured below on a book shelf, which they covered with a camouflage pattern printed on paper. As you can see in these images, real-life conditions introduce new factors to consider, namely lighting conditions and shadows.

Courtesy of MIT

Besides hiding outdoor eyesores like utility boxes and air-conditioning condensers, Owens hopes his team’s camouflaging technique could also help make interior spaces more pleasant.

“Maybe there’s an object that performs a really important function but makes the room seem cluttered, “ Owens explains in a phone interview. He says architects and designers can then analyze how people walk through and perceive the room, feed all that information to an algorithm similar to theirs, and figure out the best way to disguise that unsightly object.

 

About the Author

Most Popular

  1. photo: An elderly resident of a village in Japan's Gunma Prefecture.
    Life

    In Japan’s Vanishing Rural Towns, Newcomers Are Wanted

    Facing declining birthrates and rural depopulation, hundreds of “marginal villages” could vanish in a few decades. But some small towns are fighting back.

  2. A photo of a police officer in El Paso, Texas.
    Equity

    What New Research Says About Race and Police Shootings

    Two new studies have revived the long-running debate over how police respond to white criminal suspects versus African Americans.

  3. Transportation

    Berlin Will Spend €2 Billion Per Year to Improve Public Transit

    The German capital plans to make major investments to expand bus and rail networks, boost frequency, and get ahead of population growth. Are you jealous yet?

  4. Design

    How Advertising Conquered Urban Space

    In cities around the world, advertising is everywhere. We may try to shut it out, but it reflects who we are (or want to be) and connects us to the urban past.

  5. Environment

    Paris Wants to Grow ‘Urban Forests’ at Famous Landmarks

    The city plans to fill some small but treasured sites with trees—a climate strategy that may also change the way Paris frames its architectural heritage.

×