Seeing the out-of-sight has turned a new corner. Now, digital cameras can take an image of an object hidden around a wall, which could help autonomous cars detect hazards in blind spots.
In principle, any vertical edge can act as an accidental camera, by projecting subtle patterns of light onto the ground. These patterns reveal a semblance of what is happening on the other side of the edge and, though too faint to be noticed by the human eye, can be enhanced and interpreted by imaging algorithms.
Advertisement
Until now, however, these algorithms have required videos not images and a ground that is completely featureless.
Get rid of these unrealistic conditions and you are essentially left with an equation that has three variables and two unknowns, which is traditionally unsolvable, says Vivek Goyal at Boston University in Massachusetts in the US. Yet he and his colleagues have solved it with a new algorithm that seeks the most 鈥減lausible鈥� answer, based on the knowledge that the light patterns tend to fade as they get closer to the wall.
鈥淲e can work it all out in one snapshot,鈥� says Goyal, whose pictures of a blue cylinder taken from around a black screen were presented at the International Conference on Computational Photography in Japan this week.
Instant analysis of more realistic surfaces could help autonomous cars, which need to react quickly while driving over road markings.
Goyal鈥檚 images are currently only one-dimensional 鈥� mere lines of varying colour and intensity. But the team is working on two- and three-dimensional imaging.
