Recognition Algorithm Based On The Models Of The Vehicle Audio Signal

Vehicle classification is an important part of intelligent transport system. It is used in toll,transportation statistics and so on far and wide. Traditionally, electric cables and induction coils are buried under roads; vehicles' photos taken by camera record system. However, this method will cause damage to the road, and demand equipment maintenance for long term use. This thesis is focused on the study of the vehicle classification based on the acoustic signal.First, the present research situation of sound detection and its feasibility was reviewed. Then the generation and propagation of acoustic signal was studied. This thesis collects many kinds of signal, de-noise and synthesize specimen using the theory of wavelet and array signal. And then extract features from signal. Feature extraction is the key to the classification process. In the approach, features are extracted from spectrogram base on traditional methods. Spectrogram denotes the spectrum by way of picture in two dimensions: time and frequency. Experiments show that there are many different kinds of color and texture for power in spectrogram contrast to the power spectrum. This thesis extract the contrast,entropy and energy as our features.The last part of this study investigated pattern recognition problem over texture features. Aim at the characteristics of the signal the thesis adopt fuzzy c-means algorithm. For complex features the thesis is taken a new algorithm called Kernel-based fuzzy and possibilistic c-means clustering was taken. Some results are given to illustrate the advantages of the proposed algorithms over the FCM (fuzzy c-means) and PCM (possibilistic c-means) algorithms.The algorithm of this thesis is effective to class the vehicles base on the vehicle acoustic signal. Using the acoustic noise signal generated by driving motor vehicles to vehicle classification is a new and promising method.