A camera combined with machine learning is assisting Sandia National Laboratories and the Department of Homeland Security Science and Technology Directorate (S&T) create more precise drone detection capability than possible with visuals alone.
Acoustic, radio or thermal detection of drones always have been more reliable than visual detection. Other moving objects, such as birds or plastic bags caught in the wind, can be mistaken for drones.
For this reason, the primary method of detection has been to use acoustic or heat signatures. This method, too, is flawed as acoustic signatures tend to be less reliable in urban areas with higher levels of background noise.
“Current systems rely upon exploiting electronic signals emitted from drones,” Bryana Woo of Sandia National Laboratories, said, “but they can’t detect drones that do not transmit and receive these signals.”
Sandia is testing temporal frequency analysis that analyzes the frequency of pixel fluctuation in an image over time instead of using video or acoustic or heat signatures.
During the tests, a streaming video camera captured the impressions of three different multirotor drones moving in various directions. A machine learning algorithm analyzed the frames and rendered the full flight path of the target object in all its directions.