Research On Identification Theory Of Traveling Vehicle's State On The Basis Of Acoustic Caused By Vehicle

With the rapid development of the sensor technology and digital signal processing technology, the modern technology of monitoring vehicle plays an important role in modern traffic management system and is gradually prone to miniaturization and intelligence. A new kind of vehicle detection method connecting the vehicle sound signal and driving condition is put forward, monitoring and identifying the driving condition by analyzing the signal of moving vehicle.In the paper, the signal conditioning circuit is designed on the basis of regarding microphone as detection component and PCI-1712 card as acquiring platform,the programs of data acquisition are compiled in the programming environment of C++ builder so as to satisfy high speed data acquired.This research is based on data collected from field experiment and uses modern digital signal processing technology to analyzing, processing and simulating the vehicle's acoustic signal by Matlab7.1 software. It mainly includes pretreatment, feature extraction, classification and identification. In order to extract and highlight the useful signal, window function and advance aggravating method to process the gathering signal is used in the pretreatment. In addition, it uses methods of zero mean, normalization and wavelet de-noising to process the useful signal, and these methods are good for following research and analysis. Feature extraction mainly analyzes the signal in time domain, frequency domain, time-frequency domain and human auditory characters. The paper discusses and researches the short-time zero-crossing of signal in time domain, power spectrum in frequency domain, frequency band energy of signal in time frequency domain, and calculates MFCC from auditory characters. Finally, according to the comparison results of different analysis methods, the signal eigenvalue is extracted by the method of combining wavelet and wavelet packet. Classification and identification adopts the BP neural network. Firstly, extracted eigenvalue mentioned above is adopted as the network's input, the different driving state as the BP network's output. Then this network is trained. The trained network can be used to monitor the driving state, the output of network has effective identification to vehicle's driving condition.Through simulations and experiments, the results showed that the extracted eigenvalue using the method of combining wavelet and wavelet packet is stable, and the classifier that is designed through BP network has good recognition rate. Therefore, the paper can afford to identify different driving state by using the vehicle signal.