Using neuromorphic cameras to track quadcopters

Abstract

In recent work, we have shown that neuromorphic (event-based) cameras are highly efficient at detecting quadcopters. This is done by directly detecting the frequency of the rotating blades. This signal is highly characteristic of quadcopters, in that very few other real-world phenomena generate frequencies that are tightly clustered around a single peak. This makes this detection method highly robust to false positives, and can be generated with very little computational power. However, previous work in this direction has dealt only with detection of the presence of the drone. Here, we show that the same basic computations can also be used to localize the drone within the visual field of the camera. This allows for a system that not only alerts a user that a quadcopter is present, but also provides the extra information of where the drone is located.