音频事件检测中的多级分类研究

he development of the information age makes the technology application of the field of security to a large number of promotion,which also promotes the progress of theoretical research.The traditional video surveillance has its own limitations, so as to create a security monitoring system with audio aided decision.The accuracy and availability of the monitoring system are greatly improved with the advantages of the sound and image.In this paper, we study the audio event monitoring under the background of security, select the representative of the six events, including the door sound, the sound of the gun, the sound of footsteps, the sound of glass breaking, the sound of speaking, the sound of screaming.In the design of the classifier, this paper uses the design of two level classifier, and according to different feature sets to construct different classifiers.In addition, in the system optimization, the paper also carried out the method of smoothing, dimension reduction and reliability test, which makes the identification result of the system further improve.