That which we cannot see always makes us uneasy. Especially when we’re doing dangerous activities like driving a vehicle, we want to be able to observe our surroundings and note anything that could pose a potential threat. The reality, however, is that we are always operating with blind spots.
A team of researchers at the University of South Florida are seeking to change that with the development of a new algorithm that can generate robust, 3D images of what is behind something. In other words, it can make unseen obstacles seen to us. The team’s work, which has a wide range of implications, is published in a recent article published in Nature Communications.
The algorithm works by creating full color 3D images of what’s behind an image, ranging from a building to an oncoming car. All it takes is a single photograph to produce a reconstruction of what’s behind an object. The algorithm pulls in a significant number of visual cues and information from the single photo, such as the placement of shadows. Previous efforts to accomplish this same feat involved the use of regular cameras, leading to the generation of 2D images only.
One of the more obvious applications of this technology, which is still years away from being widely used in any capacity, is in vehicles. A noteworthy problem with many autonomous driving features of many vehicles is their ability to detect objects accurately enough to make decisions about driving patterns. The inability to see objects necessarily limits the ability of self-driving vehicles. For example, seeing around corners was a key obstacle to self-driving cars that researchers have also begun addressing. Outfitted with similar technology to the algorithm described above, self-driving vehicles may inch closer to more effective driving.
The team also notes their work could aid law enforcement, including important search and rescue efforts.
Currently, the team is working to improve the accuracy of their algorithm’s outputs, as well as how quickly they can produce results.
Sources: Science Daily; Nature Communications; Stanford
