Feasibility demonstration of acoustic surveillance drones to monitor forests

Abstract

Environmental crimes are recognized as a severe and growing global problem. This includes illegal logging and deforestation, illegal mining and trade in minerals, illegal trade and poaching of wildlife and plant. As a consequence, illegal logging alone generates an estimated value of US $30-100 billion annually[1]. Surveillance by drones is usually done using video cameras and live streaming of images. This technology is more appropriate for post-factum analysis of illegal activities than for effectively detecting such activities as they happen. In effect, the precise image analysis requires an accurate but costly real-time kinematic GPS receiver. The payload required for visual drones puts constraints on the flight range. Furthermore, illegal deforestation happens mostly at night and this makes the use of vision-based drones very challenging and expensive. Autonomous drones and acoustic technology can be used to monitor forests and compounds like harbors and warehouse areas at night for surveillance of illegal activities. Currently, acoustic technologies have not been leveraged to allow for more effective prevention of illegal activities at night based on detecting and localizing suspicious sounds. Having the novel idea of surveillance acoustic drones and their different use cases in mind, there are various challenges and open questions regarding realization of this idea. We identify the following three main types of challenges: (i) drone noise neutralization, (ii) real-time stream processing, and (iii) the design restrictions enforced by the limited payload a drone can carry. In this research, we have also considered to find innovative solutions to perform the feasibility testing of the idea using low-cost and widely-available commercial equipments. This poses the challenge of data acquisition with limited infrastructure resources. This research study addresses the aforementioned challenges by proposing a novel contribution which has two fold: (i) a simulation platform which mixes drone noise with the chainsaw sound samples together with the effect of the distance in a realistic way (ii) design and implementation of a feasible surveillance data processing pipeline for the acoustic surveillance drones in order to detect the sound of an illegal logging to deforestation. The pipeline is then evaluated and analyzed to prove the efficiency of our approach.