Developed ESMD For The Feature Extraction Of Abnormal Sound In Public Places

Currently,the countries of the world tread the safety of public places as a security strategy and video surveillance monitoring platform has been installed in different regions.Abnormal events in public places,often accompanied by screaming sound,gunfire sound,glass crackle sound,explosions and other abnormal sounds.If we can detect and identify the abnormal sound without delay,and then trigger the video surveillance system to shoot the spot on the spot,we can greatly improve the existing problems of video surveillance.Obviously,the effective monitoring of abnormal sound in public places is an effective complement to the existing public safety video surveillance.In this paper,the method of extracting abnormal sound features in public places is the key technology in this field.At present,the feature extraction of abnormal sound is based on the combination of speech processing parameters.However,most of the abnormal sound signals in public places are non speech signals such as gunfire sound,glass crackle sound,explosion sound and so on.Therefore,it is necessary to carry out the research of abnormal sound feature extraction based on non speech signal processing methods.According to the abnormal noise and background noise in public places,we proposed a developed extreme-point symmetric mode decomposition method,which was used to extract the characteristics of typical abnormal sound in public places.The experimental results show that the proposed method is effective and feasible.The main work of this paper is as follows:(1)Analysis of background noise and typical abnormal sound characteristics in public places.1)Collecting gunfire sound,screaming sound,glass crackle sound,explosion sound and different situations of background noise,then establish the abnormal sound database and background database;2)Analyzes the characteristics of four kinds of abnormal sounds,such as gunfire sound,screaming sound,glass crackle sound,explosion sound in the time domain,time frequency domain,cepstrum domain and so on,then obtains the similarities and differences of the four abnormal sound signals;3)Based on the analysis the noise of bank,ATM,retail shops,office noise in the time-frequency domain,we consider that the random distribution of t is consistent with the noise distribution model in public places.(2)This paper presents a developed extreme-point symmetric mode decomposition(D_ESMD)method for abnormal sound feature extraction in public places.1)The background noise in public places can affect the distribution of abnormal sound signals,and it can drown some extreme points of abnormal sound signals to a certain degree.When the ESMD is used to decompose the abnormal sound in the public place,all the components of the modal will contain background noise,which can affect the extraction of the abnormal sound feature,and reduce the accuracy of the recognition.This paper we propose to add a random t distribution noise sequence to the original signal,and then use the ESMD algorithm to decompose,weaken the influence of the amplitude distribution of the background noise on the abnormal sound signal.2)Using permutation entropy select the ESMD decomposition mode.Background noise results in the generation of unwanted artifacts and noise components in the decomposition of abnormal sound signals,which leads to the decrease of the recognition.For this reason,this paper proposes to add a random T distribution noise sequence to the original signal,and then use the ESMD algorithm to decompose,so as to reduce the influence of the amplitude distribution of the background noise on the abnormal sound signal.3)We proposed a developed ESMD interpolation method in this paper.When ESMD using extremum midpoint interpolation method decomposition signal,there would have some differences between the end points and the midpoint,the differences can spread from the endpoint and the midpoint to the entire signal,resulting in distortion of the decomposition mode.Therefore,this paper proposes a symmetric midpoint interpolation method to replace the extreme point interpolation,and improve the decomposition efficiency of the ESMD method on the source of the signal.