Evaluation of acoustic modality features for moving vehicle identification

In this article, the authors have studied various robust features for acoustic modality based moving vehicle identification system. Due to non-stationary signal characteristics of the vehicle and its aerodynamics, the features become prominent parameters for vehicle identification in the outdoor environment. The proposed features are extracted from the experimentally generated dataset of moving vehicles. For computation of features, various feature extraction technique especially time domain, fast Fourier transform (FFT), short-time Fourier transform (STFT) and wavelets transform (WT) are applied on the vehicle's signal. The essential features from above mentioned techniques are determined by recursive feature elimination (RFE) method. Furthermore, these proposed features are examined by Pearson correlation coefficient as well as boxplot parameters. It is observed that the Pearson correlation coefficient values for FFT based features i.e., spectral flux, MFCC-5, MFCC-7 and MFCC-13 are 0.13, 0.018, - 0.0073 and - 0.006, respectively to discriminate among the feature set of bus and gypsy. The boxplot parameters of spectral flux and MFCCs shows that there is significant variation in the experimental datasets of bus, gypsy and background noise. The SVM classifier has achieved the higher identification accuracy, 90% and better performance measures on FFT based features as compared to other feature extraction techniques for vehicle identification in the outdoor environment.

Automated summary

In this article, the authors studied various robust features for acoustic modality based moving vehicle identification system. The proposed features were extracted from the experimentally generated dataset of moving vehicles. The study showed that FFT based features had better performance in vehicle identification, with the SVM classifier achieving an identification accuracy of over 90%.