Research On Gunshot Feature Extraction And Classification In Public Places

The security of public places for social stability and harmony, people’s lives andproperty is of great significance. Currently, video surveillance in public places with thecore technology of video monitoring platform has played a positive role on reducing thecrime rate and improving the efficiency of investigation and scientific evidencecollection. Gunshot signal often contains very important information including the typeof weapon, the size of the fire and so on. This information provides first-handinformation including criminal fire intelligence and clues to solve crimes. Featureextractions of gunshot mostly follow the traditional methods of speech signal processing.Gunshot signal in public places is the typical non-speech signal with non-stationary andnonlinear. So, the study of gunshot signal provides a new train of thought for processingother types of non-speech signal. Therefore, Gunshot feature extraction andclassification in public places is of significant practical value and academic significance.In this paper, First analysis gunshot signal of public and environmental backgroundnoise characteristics, then proposed an Ensemble Empirical Mode Decomposition(EEMD) model of the gunshot signal feature extraction and recognition method andfinal results are verified through the experiment.This paper main work is as follows:①Feature analyze of the typical gunshot and background noise in public places.Several kinds of typical gunshot signal and he background noise characteristicsanalyzed such as time domain, frequency domain, frequency domain, cepstrum domainhave been done. Though statistical analysis of relevant features, this paper argues thatgunshot has common property of nonlinear, non stability and the different types of shotssuch as pistols, rifles, machine guns have more similarities than differences.Environment in public places such as square, station, wharf and other background noiseis very loud, often swamped the original gunshot signal characteristics, makes soundsignal in public places is very irregular, even destroy the original gunshot signal oftime-frequency and other related features. In one of the few studies in the literature, willbe a public place noise model is assumed as Gaussian distribution, this paper analysisassumes that the background noise in public places for symmetric stability (S S)than the Gaussian distribution is more accord with the actual situation of public places. ②Gunshot character description modeling in public places. From①to know,gunshot signal have characteristics of nonlinear, non-stability. Empirical ModeDecomposition (EMD) method is an effective method for this kind of signal. Therefore,in this paper, on the basis of the EMD, gunshot in public signal feature extraction andrecognition method is researched. Ensemble Empirical Mode Decomposition (EEMD)though adding the random Gaussian distribution sequence to analyzed signal to solvethe problem of EMD, like mode mixing. Inspired by this, the article give fullconsideration to public environment background noise distribution of the actualsituation, improve models of the EEMD put forward in the reconstructed signals useS Sdistribution instead of Gaussian distribution, EMD decomposition to thereconstructed signal, get Intrinsic Mode Function (IMF), take the overall average as thefinal IMF of original signal.③Gunshot feature extraction methods in public places. Gunshot signal after EMDdecomposition will generate a finite number of IMF components, and frequency of eachIMF component is not fixed. To this end, this article will after②in the gunshotcharacter description method proposed in this paper decomposed IMF component of thegenerated Fast Fourier Transform (FFT), according to Parseval theorem to find theenergy of the original signal and each IMF component of energy, eventually take theratio of each IMF energy to the firsthand signal energy as feature vectorExperiment results show that the proposed improved EEMD model compared withtraditional speech processing method of Mel Frequency Cepstrum Coefficient (MFCC)and Linear Prediction Cepstrum Coefficient (LPCC) characterization parameters, suchas more accurate description of gunshot. At the same time, compared with the originalEMD and EEMD our method has also a certain degree of increase, and has betterrobustness.