Real-Time Emergency Vehicle Detection using Mel Spectrograms and Regular Expressions
Authors:
Alberto Pacheco-Gonzalez,
Raymundo Torres,
Raul Chacon,
Isidro Robledo
Abstract:
In emergency situations, the high-speed movement of an ambulance through the city streets can be hindered by vehicular traffic. This work presents a method for detecting emergency vehicle sirens in real time. To obtain the audio fingerprint of a Hi-Lo siren, DSP and signal symbolization techniques were applied, which were contrasted against an audio classifier based on a deep neural network, using…
▽ More
In emergency situations, the high-speed movement of an ambulance through the city streets can be hindered by vehicular traffic. This work presents a method for detecting emergency vehicle sirens in real time. To obtain the audio fingerprint of a Hi-Lo siren, DSP and signal symbolization techniques were applied, which were contrasted against an audio classifier based on a deep neural network, using the same 280 audios of ambient sounds and 52 Hi-Lo siren audios dataset. In both methods, some classification accuracy metrics were evaluated based on its confusion matrix, resulting in the DSP algorithm having a slightly lower accuracy than the DNN model, however, it offers a self-explanatory, adjustable, portable, high performance and lower energy and consumption that makes it a more viable lower cost ADAS implementation to identify Hi-Lo sirens in real time.
△ Less
Submitted 23 June, 2024; v1 submitted 25 September, 2023;
originally announced September 2023.
