Vehicle Type Classification by Acoustic Waves with Dimension Reduction Technique

In this paper, acoustic waves radiated from the running vehicles, measured by road-side instrument, are utilized for intelligent classification of vehicle type (truck, tractor and car) based on dimension reduction. To improve the accuracy rate and real-time performance of the system, dimension reduction technique as principal component analysis (PCA) and rough set (RS) are adapted to deal with the acquired acoustic waves. Firstly, raw features are extracted from acoustic waves by Welch power spectrum estimation to get a 60-dimension feature vector. Then PCA and RS are employed respectively to remove correlations among these features, which can significantly reduce the dimension of the feature vector from 60 to 4. Finally, taking the obtained salience features as the input vector, a classifier model based on three-layered RBF neural net is constructed and applied to classify vehicle type. Experimental result shows that the presented approach is effective. Meanwhile, a comparative analysis between PCARBF model and RS-RBF model is given in terms of accuracy rate.