Environmental Intelligence for Embedded Real-time Traffic Sound Classification

Abstract

In this paper a prototype to classify sounds emitted by car engines to be used for urban traffic management in smart cities is presented. The solution is based on Artificial Neural Network (ANN) executed on a resource constrained low cost embedded micro controller integrated into a sensing unit very close to the microphone. The prototype operates without the need of being connected to a remote server through an always-on connectivity. The adoption of an on-the-edge artificial intelligence architecture brings a set of advantages: safety, reliability, promptness and low power consumption. The embedded intelligence is trained from a purpose-built dataset enclosing environmental car engine sound data. A pre trained ANN classifies sound events and counts cars approaching and leaving the microphone by sensing the Doppler Effects in the sound emitted by the moving vehicles. The results showed in this paper refer to the application running on the Sensortile Development Kit board from STMicroelectronics, featuring 128KB of RAM and 1MB of Flash memory. The development of the proposed solution was somehow inspired by the UN 2030 Agenda for Sustainable Development. The Agenda defines a plan of action for people, planet and prosperity. It does so by defining 17 Sustainable Development Goals (SDGs) to enable the transformation into a more peaceful and free society. Among them, the ones which interested the authors to develop the proposed solution were the following; Goal 3: “Ensure healthy lives and promote well-being for all at all ages” Goal 11: “Make cities and human settlements inclusive, safe, resilient and sustainable” Goal 13: “Take urgent action to combat climate change and its impacts”.