Combining remote sensing and artificial intelligence to locate sounds

Abstract

Smart cities use sensors to collect and analyze data to manage their resources more efficiently and thereby enhance the quality of life for residents, especially in densely populated cities. They monitor a wide range of information, including pollution, traffic, and parking. Urban environments feature large numbers of sounds, such as noise pollution and other specific sounds, which can be harmful to citizens. Therefore, this unrestrained growing issue of urban sounds should be addressed by smart cities to improve noise mitigation. To sidestep that problem, the “Listening system Using a Crow’s nest arraY” (LUCY) is currently in development at Fraunhofer FKIE. The acoustic system aims are to automatically detect meaningful audio events contained in noisy data, such as impulsive sounds, and to accurately estimate their geographic locations. To accomplish these tasks in near real-time, LUCY has to combine advanced array processing techniques, including beamforming, with artificial intelligence methods, such as deep learning using spectro-temporal features. The proposed acoustic system is a low-cost, small and lightweight system. It consists of a peculiar volumetric array of tiny MEMS microphones, called the “Crow’s Nest Array” (CNA), which has a crucial influence on the accuracy of the sound localization estimation, and a miniature computer to process methods including sound localization calculation. Due to its small size, LUCY can easily be deployed on numerous types of platforms, including Unmanned Aerial Vehicles (UAVs).