Research On Recognition And Location Technology Of Automobile Whistle

In recent years,the quantity of automobiles in our nation has increased quickly,and the noise of car whistle is concerned by the community.The illegally whistle on the street of the city will disturb the production and living order of the residents,especially in the nearby roads of schools,hospitals and residential areas.The state has legislated to restrict the phenomenon of illegal car whistle,and the traffic police also set up warning signs in the area where the car whistle is forbidden.For a long period,it is not easy to have a helpful way to acquire evidence for illegal car whistle,which also contributes to the fluke of the drivers.Therefore,it is urgent to install the capture equipment for illegal car whistle in the area where the whistle is forbidden.This paper explores the detection technology of car whistle through the research of sound recognition and positioning methods.The main work is as follows:First,the research on sound recognition of car whistle.In this paper,convolution neural network is introduced to recognize the car whistle.With the intention of improve the accuracy of recognition,the dataset of car whistle is established by collecting sound data.The procedure is constituted of two sections: feature extraction and recognition.In feature extraction,three features MFCC,ZCR,and STE are extracted and fused.In the recognition process,a neural network with three convolutional layers is built,then input the sound extraction features in the dataset into the network for training,so as to get a trained sound recognition model,and then use this model to recognize sounds.The results show that the model recognition precision is 95.7 %.Second,the research on accelerated operation of neural network model.In order to achieve the goal of miniaturization and embedding of car whistle detection equipment at intersections,this article studies the neural network acceleration calculation method in the embedded system.For the application scenario of car whistle recognition,the hardware device chooses the RK1808 NPU,and the model conversion and the model inference of the RK1808 NPU is studied in detail.It is verified by experiments that the NPU has better recognition accuracy and speed.Third,the research on sound source localization algorithm.Through the analysis of the microphone array model and the sound source model,for the near-field model,the sound source localization algorithm based on the time difference of arrival is studied and improved,the sound source position is calculated according to the time delay,and the near-field sound source localization is realized;For the far-field sound source model,the algorithm based on the Capon beamformer is studied,and the microphone array is designed for the Capon beamforming algorithm,and the Underbrink microphone array is established to realize the localization of the far-field sound source.Fourth,the development of sound detection equipment for car whistle.A 16-sensor Underbrink microphone array was designed,and a 16-channel sound acquisition board was designed using XMOS parallel processor.On the embedded system RK1808 NPU,a software program was developed to realize the car whistle detection function.