The aim of this research is to investigate the feasibility of developi
ng a traffic monitoring detector for the purpose of reliable on-line v
ehicle classification to aid traffic management systems. The detector
used was a directional microphone connected to a DAT (Digital Audio Ta
pe) recorder. The digital signal was preprocessed by LPC (Linear Predi
ctive Coding) parameter conversion based on autocorrelation analysis.
A Time Delay Neural Network (TDNN) was chosen to classify individual t
ravelling vehicles based on their speed-independent acoustic signature
. The paper provides a. description of the TDNN architecture and train
ing algorithm, and an overview of the LPC preprocessing and feature ex
traction technique as applied to audio monitoring of road traffic. The
performance of TDNN vehicle classification, convergence, and accuracy
for the training patterns are fully illustrated. To establish the via
bility of this classification approach, initially, recordings were car
ried out on a strip of airfield for four types of vehicles under contr
olled conditions. A TDNN network was successfully trained with 100% ac
curacy in classification for the training patterns, as well as the tes
t patterns. The net was also robust to changes in the starting positio
n of the acoustic waveforms with 86% accuracy for the same test data s
et. In the second phase of the experiment, roadside recordings were ma
de at a two-way urban road site in the city of Leeds with no control o
ver the environmental parameters such as background noise, interferenc
e from other travelling vehicles, or the speed of the recorded vehicle
. A second TDNN network was also successfully trained with 96% accurac
y for the training patterns and 84% accuracy for the test patterns. (C
) 1998 Elsevier Science Ltd. All rights reserved.