Acoustic signature-based vehicular traffic density state estimation in developing regions

Abstract

In developing regions like Asia, where the traffic conditions are chaotic and non-lane driven, the intrusive techniques may be inapplicable. The vehicular acoustic signals and the occurrence and mixture weighting of these signals are determined by the prevalent traffic density state condition. This research work considers the problem of vehicular traffic density state estimation, based on the information present in the acoustic signal acquired from roadside-installed microphone. In this work a visual analytic for consideration of frame size and shift size, while extracting feature vectors using Mel Frequency Cepstral Coefficients (MFCC) for traffic density state estimation and corresponding experimental validation is provided. Different kernel functions of support vector machine (SVM) from single acoustic frame to multiple contiguous frames were used to classify the density state as low, medium and heavy. The system results in enhanced classification performance when observed time increases or when multiple contiguous frames were considered.