Improving drone detection using acoustic methods and statistical filtering

Abstract

Acoustic methods for the detection of propeller-driven drones are used at relatively short distances, up to a few hundred meters. The drones are relatively weak sources of sound. The appearance of acoustic disturbances in the environment significantly reduces quality of the detection and localization drones by acoustic methods. Starting assumption, in resolving of this problem, is that acoustic ambient noise has a Gaussian natural distribution. A significant reduction in the influence of ambient noise was achieved by using the so-called statistical filter, which eliminates all contributions of ambient noise if their distribution is Gaussian. The function of filtering acoustic signal, that is, the filter itself, is realized as an independent module, that is, a processor that analyzes the acoustic signal using standard statistical methods, as well as higher order statistics. To check the quality of the filter itself it is used the five core quality factors: correlation coefficient, Normalized Root Mean Square Error (NRMSE), Signal-to-Error Ratio (SER), Maximum Absolute Error (MaxAE) and Estimated Delay (τ). The analyzed acoustic signals were generated in different flight mode of the drone, such as hovering, circular movement, left-right movement and others.