Research Of Vehicle Audio Features Recognition Based On Compressed Sensing Theory

It is constituted for the intelligent transportation and monitoring system by the audiofeature recognition. Therefore, recognize and judge the traffic accident, monitor real-timeintelligently using vehicle audio feature is becoming increasingly important.The audio feature is researched in this paper. By analyze and recognize five kinds ofaudio signal, it achieved the purpose of the traffic monitoring. The main idea of this paper isuse of compressed sensing theory to identify the vehicle audio features. Audio is capturedwhen the vehicle is in motion firstly. Based on short time Fourier transform andWigner-Ville Distribution, the time-frequency analysis graph is obtained by differenttime-frequency analysis methods. Through theoretical analysis and performance comparison,the best processing method is provided. The second step is to extract audio features whichare recognized in two aspects. On the one hand, the features are extracted fromWigner-Ville Distribution graph with the use of principal component analysis. On the otherhand, features include the instantaneous energy of the edge characteristics in Wigner-VilleDistribution and power spectral density. Then this paper constructs the dictionary matrix toextracted feature with the method of compressed sensing and recognizes audio.It is show that experiment use compressed sensing to recognize the vehicles audiofeature, according to its different extracted feature, the results of identification also aredifferent. The result of the features are extracted from Wigner-Ville Distribution graph withthe use of principal component analysis is better than features include the instantaneousenergy of the edge characteristics in Wigner-Ville Distribution and power spectral density.