Automatic Vehicle Classification And Recognition Based On Vehicle Acoustic Signal

Automatic vehicle classification and recognition is an important part of intelligent transport system. Various techniques such as magnestism frequency technique and video frequency technique have been used to achieve it. Using the acoustic signal generated by driving motor vehicles to achieve the traffic management is a novel and promising method.First,the theory of the acoustic signal generated by driving motor vehicles and its characteristic was summarized. And the linear model was established to model the acoustic signal generated by driving motor vehicles. Then the theory of homomorphic signal peocessing and the cepstrum was introduced. After that this study focused on the AR modeling parameter estimation and its high speed lattice formulation,based on what the linear predition coding(LPC)parameter is extracted. Then the short-time average energy,short-time average ZCT and the power spectral density were used to analyse the acoustic signal and their limitation in automatic vehicles classification and recognition was found. At last the study selected the LPCC parameter based on LPC to be the feature of the acoustic signal generated by driving motor vehicles for automatic vehicle classification and recognition,which can reflect the system response more clearly.After extracting the LPCC parameter from the acoustic signal generated by driving motor vehicles,the study used direct distance discriminance,Fisher's discriminance and K-means arithmetic to achieve the automatic vehicle classification and recognition. The study tried an experiment on four sorts of vehicles samples in automatic vehicle classification and recognition. The results proved the validity of the method.Using the acoustic signal generated by driving motor vehicles to achieve the automatic vehicle classification and recognition is a novel method. It can be used in the common road without large numbers of vehicles to examine the traffic value,or combined with other methods to make the data more veraciously.