A Virtual Fence for Drones: Efficiently Detecting Propeller Blades with a DVXplorer Event Camera

Abstract

In previous work, we prototyped a portable drone detection system using a DAVIS 346 event camera and a Raspberry Pi 4, running in 5.14 W. Here, we expand on this work by switching to the higher-resolution DVXplorer and by including a small neural network classifier system. The resulting system improves the range at which drones can be recognized (from 9m to 19m). We also demonstrate our novel in-lab test system, capable of generating controlled training data across a wide variety of lighting and optical conditions. The new 100-neuron classification system runs at 100Hz with an accuracy of 98% on our field test and 96% on the in-lab test suite.