Vehicle targets classification by acoustic signal

This paper investigated feature extraction and classification methods for vehicle targets recognition by acoustic signals. The experiment data is acquired on a bitumen road, on which there are sufficient types of vehicles and an appropriate traffic density. We extract feature vector in many ways, including MFCC, LPCC, Wavelet et al, and compare the results using Support Vector Machine as classifier. The classification of three types of vehicles is realized, which are car, motorcycle and truck. This paper also investigates the effects of the normalization of the feature vectors and the decimation of the signal on the classification accuracy. The best result is obtained from MFCC, which is over 86 percent.