16:50
Session 8: Drone Audition - Listening with Drones
Chair: Jeongwoo Ko
16:50
20 mins
|
Can Drones Hear Birds from the Sky? Hardware and Software Lessons from UAV Bioacoustic Monitoring
Lin Wang
Abstract: UAV-based bioacoustic monitoring offers a flexible way to survey birds in remote or difficult-to-access habitats, but onboard recordings are often dominated by ego-noise from propellers, motors, airflow and flight manoeuvres. This paper summarises our recent three studies on UAV bioacoustic monitoring, progressing from hardware feasibility for bird surveys, to drone audition with combined hardware and software solutions, and finally to in-situ aerial monitoring from extremely noisy drone recordings. The studies show that microphone placement, directional recording, wind protection, suspended payload design, noise-augmented learning and deep-learning enhancement can all improve the recovery of bird vocalisations from noisy UAV audio. Together, these studies highlight the importance of hardware-software co-design for UAV bioacoustic monitoring under severe ego-noise.
|
17:10
20 mins
|
Listening with drones: Denoising algorithms for best performance
Paula Chale Castell, Roberto Merino-Martinez, Toros Senan, Stephy Annie Curie
Abstract: Unmanned Aerial Vehicles (UAVs) have emerged as indispensable assets in modern search-and-rescue operations, providing rapid situational awareness in environments where ground access is restricted. While optical and thermal sensors are standard, their effectiveness is often compromised by occlusion, dense vegetation, or adverse visual conditions, such as smoke and darkness. Drone audition, i.e. the use of embedded microphone arrays to detect and localize victims, offers a promising complementary modality. However, the practical deployment of this technology is currently hindered by severe acoustic interference from the UAV’s propulsion system, typically known as ego-noise, which drastically degrades the Signal-to-Noise Ratio (SNR) and masks target signals.
This research presents a comprehensive framework for evaluating and enhancing acoustic detection from aerial platforms. The methodology adopts a rigorous, two-stage validation approach designed to bridge the gap between theoretical algorithms and real-world applicability. In the first phase, we conduct a comparative benchmark of diverse noise suppression and Direction-of-Arrival (DoA) estimation architectures using previously recorded acoustic datasets. This establishes a performance baseline against ground-truth data, allowing for the optimization of filter parameters prior to hardware implementation.
Subsequently, the study transitions to experimental validation using a microphone array and a DJI Mavic 3 Enterprise drone. These experiments are conducted within the anechoic chamber of Delft University of Technology to simulate ideal free-field sound propagation. This controlled environment effectively isolates the system from environmental reverberations and multipath effects, ensuring that acoustic interference is exclusively attributable to the drone’s ego-noise profile. By systematically analyzing the trade-offs between localization accuracy, noise rejection capabilities, and computational cost across different processing strategies, this study aims to identify robust acoustic sensing configurations suitable for integration into autonomous rescue systems.
|
|