Design of an algorithm for acoustic signal detection of moving vehicles

TL;DR

This paper presents a novel algorithm inspired by LOG filtering for detecting and counting moving vehicles using acoustic signals, providing a scalable and validated method for traffic monitoring.

Contribution

It introduces a new acoustic vehicle detection algorithm based on second derivative analysis, adapted from image processing techniques, for the first time in one-dimensional signals.

Findings

Effective vehicle detection and counting in acoustic signals

Algorithm is simple, scalable, and validated with real data

Potential application in urban traffic management

Abstract

The precise detection and counting of vehicles is fundamental to know the state of agglomeration of the roads, as well as to anticipate possible actions of regulation of traffic or future changes that it is necessary to carry out in the urban design in the management of the transport. It is in fact part of the programs of many municipalities for statistical purposes, for the planning, design, extension and modeling of structures. The operator L.O.G. (Laplacian of gaussian) is commonly used for the detection of elements in images, by the analysis of its border, and is a widely extended method in image filtering. However, as such is not an extended procedure in the analysis of one-dimensional signals. We developed a procedure inspired by LOG filtering, and performed the analysis of the acoustic signal obtained by a portable recorder located on the side of a single lane road, by evaluating the negative minima of the second discrete derivative of the smoothed acoustic signal, filtering the selection with a threshold, in order to detect and count the passage of vehicles along the road. We obtain a simple procedure, scalable, and adequately validated by the results for that purpose.