Seeing what is happening on the other side of a corner isn't as hopeless as it seems. Scientists are working on the problem for decades, using lasers to bounce off light hidden objects and discover what is moving outside of their line of sight. Now, researchers from MIT's CSAIL have moved one step further: they're using footage from an ordinary smartphone to "see" around corners by seeing subtle fluctuations in light and darkness.
The assumption of the work is simple: all objects reflect light, and, by carefully analyzing the floor near a corner, so you can see if something is shifting on the other side based on altering darkness. These changes are imperceptible to the human eye, however, scientists were able to place them by tweaking the footage from normal industrial cameras, and even an iPhone 5s.
The procedure has several serious limitations, however. For starters, you can not make out any detail regarding the hidden object. You can identify how fast it's moving, and get some idea of its own position, however you can't make out edges or shape or texture. And, unlike the laser way of viewing around corners, then the hidden individual or thing has to be moving in a brightly lit space so as to be discovered. As a previous limitation, the origin footage also has to be steady, although the researchers are still working on how to use moving footage.
These caveats aside, the system is really robust. It works fine outside in bright sunshine, as an example, and can even be utilised in the rain. Its creators think it could be utilised later on for cameras to self-driving cars, permitting them to look around corners to identify pedestrians, cyclists, as well as other vehicles. In that type of situation that you don't need to observe some detail, so you only need to know if something is there.
Lead author of the research, Katie Bouman, claims that the MIT CSAIL team even analyzed exactly this kind of scenario. In order to see how feasible the method would be for self-driving cars in our paper, we took a video when we were pretty far away from the corner at a shallow angle, said Bouman.
The machine -- called CornerCameras -- now wants a laptop to perform the necessary image processing, however Bouman claims this might be overcome later on. From a computing point of view, I think our system could be fully put on a phone. We just haven't done it [yet], she told. Just wait till there is an app for it.