Cyclist Safety Device based on Spectral Audio Processing for Approaching Vehicles Detection

Safety is one of the key points that limit the spread of light mobility. The objective of this work is the proposal of a proof of concept of an embedded, low-cost, battery-powered system to be placed on the back of the bike capable of recognizing an approaching car in advance and accessing a visual signal that warns the coming car of the cyclist's presence. Many studies in the literature on vehicle approach detection use audio-video techniques; however, they unsatisfactorily apply to embedded implementation due to their computational complexity. Based on the analysis of a publicly available data-set of audio recordings, this study proposes a technique that can detect the car approaching by low-frequency spectral audio signal analysis. The real-time acquired signal, while the bike is in motion, is compared to the detection of the background spectral noise obtained in static conditions of the bike.