Acoustic Drone Package Delivery Detection

TL;DR

This work introduces the first acoustic system for detecting drone-delivery events using a ground-based microphone array. It employs a multitask convolutional-recurrent neural network to estimate the blade passing frequencies $BPF = N_p \omega$ for the two fastest motors and a drone activity indicator from short-time acoustic features, followed by a change-point based delivery detector that flags delivery moments from shifts in the estimated BPF distributions. Results show the BPF estimator achieving a mean absolute error around 6 Hz for onboard tests and robust performance against substantial noise, while the delivery-detection pipeline identifies 96% of delivery frames at an 8% false-positive rate, with practical latency considerations. The study demonstrates feasibility up to 100 m and highlights limitations, including domain generalization to multiple drone models and the need to incorporate Doppler and maneuver effects for robust real-world deployment.

Abstract

In recent years, the illicit use of unmanned aerial vehicles (UAVs) for deliveries in restricted area such as prisons became a significant security challenge. While numerous studies have focused on UAV detection or localization, little attention has been given to delivery events identification. This study presents the first acoustic package delivery detection algorithm using a ground-based microphone array. The proposed method estimates both the drone's propeller speed and the delivery event using solely acoustic features. A deep neural network detects the presence of a drone and estimates the propeller's rotation speed or blade passing frequency (BPF) from a mel spectrogram. The algorithm analyzes the BPFs to identify probable delivery moments based on sudden changes before and after a specific time. Results demonstrate a mean absolute error of the blade passing frequency estimator of 16 Hz when the drone is less than 150 meters away from the microphone array. The drone presence detection estimator has a accuracy of 97%. The delivery detection algorithm correctly identifies 96% of events with a false positive rate of 8%. This study shows that deliveries can be identified using acoustic signals up to a range of 100 meters.

Figures (8)

• Figure 1: BPF distribution before and after a delivery. A payload shifts the distribution to the right as the effective drone weight increases. The ground truth BPF data from multiple delivery segments were used to generate this graph resulting in approximately 25 000 data points.
• Figure 2: Architecture overview of the proposed algorithm.
• Figure 3: Delivery detection algorithm overview. $D_{\chi^2}$ is the Chi squared distance, $D_{JSD}$ the Jensen–Shannon divergence, $D_{HI}$ the intersection score and $\Delta\mu_{a,b}$ the difference in average between both histograms
• Figure 4: The modified Holybro X500 V2 drone and the 16 microphone array
• Figure 5: Measured BPF for PWM signal ranging from 1100 us (10% throttle) to 2000 us (100% throttle)