Design Of Abnormal Sound Event Monitoring System In Public Place Based On Edge Computing

At present,video technology is often used to monitor public places,and there are problems such as fixed scope of monitoring and easy to block the monitoring field of vision.In addition,video surveillance is not capable of monitoring abnormal sound events that need rapid identification and location,such as explosion and alarm.In order to solve the above problems,this paper studies related algorithms and programming implementation from the perspective of sound source location and abnormal sound event classification,and uses edge computing to build a set of abnormal sound event monitoring system in public places.In terms of sound source location,in order to locate the sound source in the public environment containing noise and reverberation,this paper uses SRP-PHAT(Steered Response Power using Phase Transform)as the sound source location algorithm used by the monitoring system.Limited by the performance of edge computing equipment,the space search part of SRP-PHAT algorithm is improved to reduce the computational cost.The simulation results show that the computational cost of the improved SRPPHAT algorithm is reduced while maintaining a certain sound source location performance.In addition,because the SRP-PHAT algorithm can only obtain the azimuth and pitch angle of the sound source relative to the microphone array,this paper uses the joint positioning method of multiple microphones to achieve accurate positioning of the sound source in three-dimensional space,and verifies its positioning performance in the simulation experiment.In terms of abnormal sound event classification,this paper uses deep learning technology to build a convolutional neural network to classify the detected sound events.Meanwhile,CNN6 network with fewer convolution layers is used in this paper for edge devices with limited GPU resources,and model compression methods such as channel pruning and knowledge distillation are applied to it.Experimental results show that although the m AP value of CNN6 network decreases by 9.8% after model compression,the parameter number of CNN6 network decreases by 70.8%,which can be applied in edge computing devices.Finally,Jetson Nano is used as edge computing device to complete the design and construction of the monitoring system by comprehensively utilizing the research results of sound source location and abnormal sound event classification.The test results show that the monitoring system designed in this paper has strong ability of sound source location and sound event classification,which can provide reference for the development of related systems.