When an unauthorized drone is being flown in a restricted airspace, the authorities understandably want to locate its operator. A new AI-based system may allow them to do so, succeeding where other technologies fail.
First of all, it is possible to determine the approximate location of a drone's pilot, using multiple widely-spaced sensors to triangulate the originating point of its radio control signal. Those sensors do already have to be in place, however, plus they may not be able to pick up the radio signal if it's obstructed by other wireless signals (such as those from Wi-Fi or Bluetooth) that are present in the area.
Seeking a better alternative, researchers from Israel's Ben-Gurion University of the Negev have developed a system in which cameras optically track the drone's flight path in three-dimensional space. That video is analyzed utilizing a deep neural network, which was "trained" on previously-recorded footage in which the location of the drone operator was known.
Since pilots typically keep their drones within line of sight (or if nothing else, within radio range), analysis of flight patterns such as the aircraft's changing vertical and horizontal location, along with its tendency to move along a path that arcs around a central point that's off to one side, can be used to ascertain where the pilot is.
As a result, in computer simulations utilized so far, the system was able to locate a drone's operator with 78-percent accuracy. That figure should rise as the technology is developed further, which will include testing it on actual drones in the real world.
A paper on the research, which is being led by computer science student Eliyahu Mashhadi, was recently presented at the online Fourth International Symposium on Cyber Security, Cryptography and Machine Learning.
Source: American Associates, Ben-Gurion University of the Negev via EurekAlert
