(POSTER) Drone Detection and Localization Using Low-Cost Microphone Arrays and Convolutional Neural Networks

This paper examines the possibility of using low-cost commercial off-the-shelf audio recording equipment in combination with machine learning techniques to discover the presence of hostile UAVs. A convolutional neural network (CNN) was trained to detect and localize two types of quadrotor drones using ground truth position data collected with motion capture equipment. System performance was evaluated on pre-recorded validation data sets and in realtime operation. In both cases, drones can be successfully detected and localized within the constrained working volumes studied, achieving angular accuracies in the 8–13° range. However, further work remains to be done before system feasibility in outdoor conditions can be established.