WE MAY soon be able to see what is lurking out of sight, thanks to an algorithm that allows a camera to make out objects hidden from view. This might one day enable autonomous vehicles to spot hazards before they are visible.
The algorithm was created by Vivek Goyal at Boston University in Massachusetts and colleagues. It works by analysing the faint reflections of obscured objects on a surface such as a wall. Specialist equipment isn鈥檛 needed to gather the images of reflections 鈥� a normal digital camera suffices.
Advertisement
To test the system, an LCD monitor was placed 1 metre away from a white wall. The team then placed a digital camera so that it also faced the white wall, but with an opaque dividing screen between the camera and monitor. Different images were displayed by the monitor and the algorithm attempted to recreate them using only the reflection.
The results, however, were a little disappointing, so Goyal and his colleagues tweaked the set-up so that a small square object, the size of two credit cards, stood between the monitor and wall. This blocked some of the reflection in a predictable way, giving the algorithm extra information about the hidden object (Nature, ).
This greatly improved the reconstructions, which although pixelated, accurately showed features of the original images on the screen, including colours and shapes (see pictures below).
鈥淲e are able to use natural illumination and a simple camera to reconstruct images that aren鈥檛 within direct sight,鈥� says Goyal.
Other groups have managed to reconstruct objects from their reflections, but only by using lasers and costly sensors. 鈥淭here are situations where you can鈥檛 shine laser light everywhere and still be stealthy,鈥� says Goyal.
The team thinks that phone cameras may be good enough for the technique and that eventually they will be able to create an app for seeing objects around corners.
鈥淭he results are stunning,鈥� says Gordon Wetzstein at Stanford University, California. He says this is the first time that a technique like this has produced full-colour reconstructions.
The approach could prove to be helpful for autonomous vehicles, allowing them to see other traffic before it goes around a bend.
