Acoustic Based Vehicular Traffic Density Estimation

In this study, traffic density is estimated using acoustic noise signals formed by the land vehicles. The acoustic noise signals formed by the vehicles consist of engine noise, air turbulence, the noise of the wheels touching the floor, exhaust noise and the horn noise. The contributions of these different types of noise change according to the traffic density. For example, engine noise and horn noise are dense when the traffic is busy and when the traffic is free-flow, air turbulence and wheel noise are more dense. By taking advantage of this change in the acoustic noise signal, the traffic density is categorized into three classes; busy, normal and free-flow. The proposed method use Mel-Frequency Cepstral Coefficients (MFCC) to extract features and the k-Nearest Neighbor Rule to classify. A data set was formed on E5 roadway and it was used to evaluate the proposed method. The effect of MFCC attributes on the traffic density estimation was investigated and the number of cepstral coefficients and the duration of windows are found to be the most important ones. It is shown that the performance of the traffic density estimation is increased if the weather conditions are considered when training the classifiers. The reason behind this improvement is investigated and shown on a two dimensional feature space. The traffic density in the E5 roadway is determined by %90 and %82 accuracies when raining and not raining, respectively