In field tests, a quadcopter was slightly better than humans at finding and following a previously-unseen trail
(Credit: University of Zurich)
By BEN COXWORTH FEBRUARY 10, 2016
It's becoming increasingly likely that in the not-too-distant future, a robot may be what finds you if you're trapped in rubble at a disaster site. Now, it's also looking like a drone might come to your aid if you should get lost in the woods. That's because scientists have developed machine learning-based software that already allows quadcopters to follow forest paths better than humans.
The program was created by researchers at the University of Zurich, the Università della Svizzera italiana, and the University of Applied Sciences and Arts of Southern Switzerland.
In the course of its development, team members spent several hours hiking along trails in the Swiss Alps, taking over 20,000 photos with a helmet-mounted camera as they did so. The software analyzed these images, using a deep neural network to teach itself the distinct (and sometimes subtle) features that differentiate a trail from the surrounding environment.
The system was then used in a quadcopter, which was equipped with two video cameras for stereoscopic computer vision. When placed on a previously-unseen trail, the drone was able to autonomously orient itself and follow the trail with an 85 percent accuracy rate – humans, by contrast, scored 82 percent.
It is hoped that eventually, multiple drones could be combined with human search-and-rescue teams, allowing a greater area to be covered within the same amount of time. Additionally, the aircraft could be used to check hazardous trails, minimizing the risk to human searchers.
Before that can happen, however, the system needs to be developed further. In its current form, for instance, it's still not capable of identifying objects as being humans when it finds them.
A paper on the research was recently published in the journal IEEE Robotics and Automation Letters.
Source: University of Zurich
Full article here